spaCy/spacy/syntax/arc_eager.pyx

607 lines
21 KiB
Cython
Raw Normal View History

# cython: profile=True
# cython: cdivision=True
# cython: infer_types=True
2017-04-15 14:05:15 +03:00
# coding: utf-8
from __future__ import unicode_literals
2017-10-27 20:45:57 +03:00
from cpython.ref cimport Py_INCREF
2015-06-09 02:41:09 +03:00
from cymem.cymem cimport Pool
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
from collections import OrderedDict, defaultdict, Counter
from thinc.extra.search cimport Beam
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
import json
2017-04-15 14:05:15 +03:00
from .stateclass cimport StateClass
2017-10-27 20:45:57 +03:00
from ._state cimport StateC
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
from . import nonproj
2017-04-15 14:05:15 +03:00
from .transition_system cimport move_cost_func_t, label_cost_func_t
2017-10-27 20:45:57 +03:00
from ..gold cimport GoldParse, GoldParseC
2017-04-15 14:05:15 +03:00
from ..structs cimport TokenC
2015-06-09 02:41:09 +03:00
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
# Calculate cost as gold/not gold. We don't use scalar value anyway.
cdef int BINARY_COSTS = 1
DEF NON_MONOTONIC = True
DEF USE_BREAK = True
2014-12-16 14:44:43 +03:00
cdef weight_t MIN_SCORE = -90000
2014-12-16 14:44:43 +03:00
# Break transition from here
# http://www.aclweb.org/anthology/P13-1074
2014-12-16 14:44:43 +03:00
cdef enum:
SHIFT
REDUCE
LEFT
RIGHT
2015-05-11 17:12:03 +03:00
BREAK
2015-05-11 17:12:03 +03:00
2014-12-16 14:44:43 +03:00
N_MOVES
2015-05-11 17:12:03 +03:00
MOVE_NAMES = [None] * N_MOVES
MOVE_NAMES[SHIFT] = 'S'
MOVE_NAMES[REDUCE] = 'D'
MOVE_NAMES[LEFT] = 'L'
MOVE_NAMES[RIGHT] = 'R'
MOVE_NAMES[BREAK] = 'B'
2015-01-27 19:18:43 +03:00
# Helper functions for the arc-eager oracle
cdef weight_t push_cost(StateClass stcls, const GoldParseC* gold, int target) nogil:
cdef weight_t cost = 0
cdef int i, S_i
for i in range(stcls.stack_depth()):
S_i = stcls.S(i)
if gold.heads[target] == S_i:
cost += 1
if gold.heads[S_i] == target and (NON_MONOTONIC or not stcls.has_head(S_i)):
cost += 1
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
if BINARY_COSTS and cost >= 1:
return cost
2017-05-30 21:37:24 +03:00
cost += Break.is_valid(stcls.c, 0) and Break.move_cost(stcls, gold) == 0
return cost
cdef weight_t pop_cost(StateClass stcls, const GoldParseC* gold, int target) nogil:
cdef weight_t cost = 0
cdef int i, B_i
for i in range(stcls.buffer_length()):
B_i = stcls.B(i)
cost += gold.heads[B_i] == target
cost += gold.heads[target] == B_i
if gold.heads[B_i] == B_i or gold.heads[B_i] < target:
break
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
if BINARY_COSTS and cost >= 1:
return cost
2017-05-30 21:37:24 +03:00
if Break.is_valid(stcls.c, 0) and Break.move_cost(stcls, gold) == 0:
cost += 1
return cost
cdef weight_t arc_cost(StateClass stcls, const GoldParseC* gold, int head, int child) nogil:
if arc_is_gold(gold, head, child):
return 0
elif stcls.H(child) == gold.heads[child]:
return 1
# Head in buffer
2017-05-30 21:37:24 +03:00
elif gold.heads[child] >= stcls.B(0) and stcls.B(1) != 0:
return 1
else:
return 0
2015-06-10 07:56:35 +03:00
cdef bint arc_is_gold(const GoldParseC* gold, int head, int child) nogil:
2017-05-30 21:37:24 +03:00
if not gold.has_dep[child]:
return True
elif gold.heads[child] == head:
return True
else:
return False
2017-05-28 15:06:40 +03:00
cdef bint label_is_gold(const GoldParseC* gold, int head, int child, attr_t label) nogil:
2017-05-30 21:37:24 +03:00
if not gold.has_dep[child]:
return True
2017-05-30 21:37:24 +03:00
elif label == 0:
return True
elif gold.labels[child] == label:
return True
else:
return False
2015-06-10 07:56:35 +03:00
cdef bint _is_gold_root(const GoldParseC* gold, int word) nogil:
2017-05-30 21:37:24 +03:00
return gold.heads[word] == word or not gold.has_dep[word]
cdef class Shift:
@staticmethod
2017-05-28 15:06:40 +03:00
cdef bint is_valid(const StateC* st, attr_t label) nogil:
sent_start = st._sent[st.B_(0).l_edge].sent_start
return st.buffer_length() >= 2 and not st.shifted[st.B(0)] and sent_start != 1
2015-06-09 02:41:09 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef int transition(StateC* st, attr_t label) nogil:
st.push()
st.fast_forward()
2014-12-16 14:44:43 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef weight_t cost(StateClass st, const GoldParseC* gold, attr_t label) nogil:
return Shift.move_cost(st, gold) + Shift.label_cost(st, gold, label)
@staticmethod
cdef inline weight_t move_cost(StateClass s, const GoldParseC* gold) nogil:
return push_cost(s, gold, s.B(0))
@staticmethod
2017-05-28 15:06:40 +03:00
cdef inline weight_t label_cost(StateClass s, const GoldParseC* gold, attr_t label) nogil:
return 0
cdef class Reduce:
2015-06-09 02:41:09 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef bint is_valid(const StateC* st, attr_t label) nogil:
return st.stack_depth() >= 2
2015-06-09 02:41:09 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef int transition(StateC* st, attr_t label) nogil:
if st.has_head(st.S(0)):
st.pop()
else:
st.unshift()
st.fast_forward()
@staticmethod
2017-05-28 15:06:40 +03:00
cdef weight_t cost(StateClass s, const GoldParseC* gold, attr_t label) nogil:
return Reduce.move_cost(s, gold) + Reduce.label_cost(s, gold, label)
@staticmethod
cdef inline weight_t move_cost(StateClass st, const GoldParseC* gold) nogil:
cost = pop_cost(st, gold, st.S(0))
if not st.has_head(st.S(0)):
# Decrement cost for the arcs e save
for i in range(1, st.stack_depth()):
S_i = st.S(i)
if gold.heads[st.S(0)] == S_i:
cost -= 1
if gold.heads[S_i] == st.S(0):
cost -= 1
2017-05-30 21:37:24 +03:00
if Break.is_valid(st.c, 0) and Break.move_cost(st, gold) == 0:
cost -= 1
return cost
@staticmethod
2017-05-28 15:06:40 +03:00
cdef inline weight_t label_cost(StateClass s, const GoldParseC* gold, attr_t label) nogil:
return 0
cdef class LeftArc:
2015-06-09 02:41:09 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef bint is_valid(const StateC* st, attr_t label) nogil:
sent_start = st._sent[st.B_(0).l_edge].sent_start
return sent_start != 1
2015-06-09 02:41:09 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef int transition(StateC* st, attr_t label) nogil:
st.add_arc(st.B(0), st.S(0), label)
st.pop()
st.fast_forward()
@staticmethod
2017-05-28 15:06:40 +03:00
cdef weight_t cost(StateClass s, const GoldParseC* gold, attr_t label) nogil:
return LeftArc.move_cost(s, gold) + LeftArc.label_cost(s, gold, label)
@staticmethod
cdef inline weight_t move_cost(StateClass s, const GoldParseC* gold) nogil:
cdef weight_t cost = 0
if arc_is_gold(gold, s.B(0), s.S(0)):
# Have a negative cost if we 'recover' from the wrong dependency
return 0 if not s.has_head(s.S(0)) else -1
else:
# Account for deps we might lose between S0 and stack
if not s.has_head(s.S(0)):
for i in range(1, s.stack_depth()):
cost += gold.heads[s.S(i)] == s.S(0)
cost += gold.heads[s.S(0)] == s.S(i)
return cost + pop_cost(s, gold, s.S(0)) + arc_cost(s, gold, s.B(0), s.S(0))
2014-12-16 14:44:43 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef inline weight_t label_cost(StateClass s, const GoldParseC* gold, attr_t label) nogil:
return arc_is_gold(gold, s.B(0), s.S(0)) and not label_is_gold(gold, s.B(0), s.S(0), label)
cdef class RightArc:
@staticmethod
2017-05-28 15:06:40 +03:00
cdef bint is_valid(const StateC* st, attr_t label) nogil:
# If there's (perhaps partial) parse pre-set, don't allow cycle.
sent_start = st._sent[st.B_(0).l_edge].sent_start
return sent_start != 1 and st.H(st.S(0)) != st.B(0)
2015-06-09 02:41:09 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef int transition(StateC* st, attr_t label) nogil:
st.add_arc(st.S(0), st.B(0), label)
st.push()
st.fast_forward()
@staticmethod
2017-05-28 15:06:40 +03:00
cdef inline weight_t cost(StateClass s, const GoldParseC* gold, attr_t label) nogil:
return RightArc.move_cost(s, gold) + RightArc.label_cost(s, gold, label)
@staticmethod
cdef inline weight_t move_cost(StateClass s, const GoldParseC* gold) nogil:
if arc_is_gold(gold, s.S(0), s.B(0)):
return 0
2016-02-01 04:37:08 +03:00
elif s.c.shifted[s.B(0)]:
return push_cost(s, gold, s.B(0))
else:
return push_cost(s, gold, s.B(0)) + arc_cost(s, gold, s.S(0), s.B(0))
@staticmethod
2017-05-28 15:06:40 +03:00
cdef weight_t label_cost(StateClass s, const GoldParseC* gold, attr_t label) nogil:
return arc_is_gold(gold, s.S(0), s.B(0)) and not label_is_gold(gold, s.S(0), s.B(0), label)
cdef class Break:
2015-06-09 02:41:09 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef bint is_valid(const StateC* st, attr_t label) nogil:
2015-06-09 02:41:09 +03:00
cdef int i
if not USE_BREAK:
return False
elif st.at_break():
2015-06-09 02:41:09 +03:00
return False
elif st.stack_depth() < 1:
return False
elif st.B_(0).l_edge < 0:
return False
2017-10-09 01:02:45 +03:00
elif st._sent[st.B_(0).l_edge].sent_start < 0:
return False
2015-06-09 02:41:09 +03:00
else:
2017-08-26 03:50:55 +03:00
return True
2015-06-09 02:41:09 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef int transition(StateC* st, attr_t label) nogil:
st.set_break(st.B_(0).l_edge)
st.fast_forward()
@staticmethod
2017-05-28 15:06:40 +03:00
cdef weight_t cost(StateClass s, const GoldParseC* gold, attr_t label) nogil:
return Break.move_cost(s, gold) + Break.label_cost(s, gold, label)
@staticmethod
cdef inline weight_t move_cost(StateClass s, const GoldParseC* gold) nogil:
cdef weight_t cost = 0
cdef int i, j, S_i, B_i
for i in range(s.stack_depth()):
S_i = s.S(i)
for j in range(s.buffer_length()):
B_i = s.B(j)
cost += gold.heads[S_i] == B_i
cost += gold.heads[B_i] == S_i
if cost != 0:
return cost
# Check for sentence boundary --- if it's here, we can't have any deps
# between stack and buffer, so rest of action is irrelevant.
s0_root = _get_root(s.S(0), gold)
b0_root = _get_root(s.B(0), gold)
if s0_root != b0_root or s0_root == -1 or b0_root == -1:
return cost
else:
return cost + 1
2015-08-09 00:31:54 +03:00
@staticmethod
2017-05-28 15:06:40 +03:00
cdef inline weight_t label_cost(StateClass s, const GoldParseC* gold, attr_t label) nogil:
return 0
cdef int _get_root(int word, const GoldParseC* gold) nogil:
2017-08-26 03:50:55 +03:00
while gold.heads[word] != word and gold.has_dep[word] and word >= 0:
word = gold.heads[word]
2017-05-30 21:37:24 +03:00
if not gold.has_dep[word]:
return -1
else:
return word
2015-08-09 00:31:54 +03:00
2017-03-11 20:12:01 +03:00
cdef void* _init_state(Pool mem, int length, void* tokens) except NULL:
2017-11-14 04:11:40 +03:00
st = new StateC(<const TokenC*>tokens, length)
for i in range(st.length):
if st._sent[i].dep == 0:
st._sent[i].l_edge = i
st._sent[i].r_edge = i
st._sent[i].head = 0
st._sent[i].dep = 0
st._sent[i].l_kids = 0
st._sent[i].r_kids = 0
2017-03-11 20:12:01 +03:00
st.fast_forward()
return <void*>st
cdef class ArcEager(TransitionSystem):
2017-03-11 20:12:01 +03:00
def __init__(self, *args, **kwargs):
TransitionSystem.__init__(self, *args, **kwargs)
self.init_beam_state = _init_state
@classmethod
def get_actions(cls, **kwargs):
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
min_freq = kwargs.get('min_freq', None)
actions = defaultdict(lambda: Counter())
actions[SHIFT][''] = 1
actions[REDUCE][''] = 1
for label in kwargs.get('left_labels', []):
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
actions[LEFT][label] = 1
actions[SHIFT][label] = 1
for label in kwargs.get('right_labels', []):
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
actions[RIGHT][label] = 1
actions[REDUCE][label] = 1
for raw_text, sents in kwargs.get('gold_parses', []):
for (ids, words, tags, heads, labels, iob), ctnts in sents:
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
heads, labels = nonproj.projectivize(heads, labels)
for child, head, label in zip(ids, heads, labels):
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
if label.upper() == 'ROOT' :
label = 'ROOT'
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
if head == child:
actions[BREAK][label] += 1
elif head < child:
actions[RIGHT][label] += 1
actions[REDUCE][''] += 1
elif head > child:
actions[LEFT][label] += 1
actions[SHIFT][''] += 1
if min_freq is not None:
for action, label_freqs in actions.items():
for label, freq in list(label_freqs.items()):
if freq < min_freq:
label_freqs.pop(label)
# Ensure these actions are present
actions[BREAK].setdefault('ROOT', 0)
actions[RIGHT].setdefault('subtok', 0)
actions[LEFT].setdefault('subtok', 0)
# Used for backoff
actions[RIGHT].setdefault('dep', 0)
actions[LEFT].setdefault('dep', 0)
return actions
property action_types:
def __get__(self):
return (SHIFT, REDUCE, LEFT, RIGHT, BREAK)
def is_gold_parse(self, StateClass state, GoldParse gold):
predicted = set()
truth = set()
for i in range(gold.length):
if gold.cand_to_gold[i] is None:
continue
if state.safe_get(i).dep:
2017-10-27 20:45:57 +03:00
predicted.add((i, state.H(i),
self.strings[state.safe_get(i).dep]))
else:
predicted.add((i, state.H(i), 'ROOT'))
id_, word, tag, head, dep, ner = gold.orig_annot[gold.cand_to_gold[i]]
truth.add((id_, head, dep))
return truth == predicted
2017-05-26 19:31:23 +03:00
def has_gold(self, GoldParse gold, start=0, end=None):
end = end or len(gold.heads)
if all([tag is None for tag in gold.heads[start:end]]):
return False
else:
return True
def preprocess_gold(self, GoldParse gold):
2017-05-26 19:31:23 +03:00
if not self.has_gold(gold):
return None
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
for i, (head, dep) in enumerate(zip(gold.heads, gold.labels)):
2017-10-27 20:45:57 +03:00
# Missing values
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
if head is None or dep is None:
gold.c.heads[i] = i
2017-05-30 21:37:24 +03:00
gold.c.has_dep[i] = False
else:
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
if head > i:
action = LEFT
elif head < i:
action = RIGHT
else:
action = BREAK
if dep not in self.labels[action]:
if action == BREAK:
dep = 'ROOT'
elif nonproj.is_decorated(dep):
backoff = nonproj.decompose(dep)[0]
if backoff in self.labels[action]:
dep = backoff
else:
dep = 'dep'
else:
dep = 'dep'
2017-05-30 21:37:24 +03:00
gold.c.has_dep[i] = True
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
if dep.upper() == 'ROOT':
dep = 'ROOT'
gold.c.heads[i] = head
gold.c.labels[i] = self.strings.add(dep)
return gold
def get_beam_parses(self, Beam beam):
parses = []
probs = beam.probs
for i in range(beam.size):
state = <StateC*>beam.at(i)
if state.is_final():
self.finalize_state(state)
prob = probs[i]
parse = []
for j in range(state.length):
head = state.H(j)
label = self.strings[state._sent[j].dep]
parse.append((head, j, label))
parses.append((prob, parse))
return parses
cdef Transition lookup_transition(self, object name) except *:
if '-' in name:
move_str, label_str = name.split('-', 1)
2015-08-09 00:31:54 +03:00
label = self.strings[label_str]
else:
2015-08-09 00:31:54 +03:00
move_str = name
label = 0
move = MOVE_NAMES.index(move_str)
for i in range(self.n_moves):
if self.c[i].move == move and self.c[i].label == label:
return self.c[i]
return Transition(clas=0, move=MISSING, label=0)
2017-05-28 15:06:40 +03:00
def move_name(self, int move, attr_t label):
label_str = self.strings[label]
if label_str:
return MOVE_NAMES[move] + '-' + label_str
else:
return MOVE_NAMES[move]
2017-05-28 15:06:40 +03:00
cdef Transition init_transition(self, int clas, int move, attr_t label) except *:
# TODO: Apparent Cython bug here when we try to use the Transition()
# constructor with the function pointers
cdef Transition t
t.score = 0
t.clas = clas
t.move = move
t.label = label
if move == SHIFT:
t.is_valid = Shift.is_valid
t.do = Shift.transition
t.get_cost = Shift.cost
elif move == REDUCE:
t.is_valid = Reduce.is_valid
t.do = Reduce.transition
t.get_cost = Reduce.cost
elif move == LEFT:
t.is_valid = LeftArc.is_valid
t.do = LeftArc.transition
t.get_cost = LeftArc.cost
elif move == RIGHT:
t.is_valid = RightArc.is_valid
t.do = RightArc.transition
t.get_cost = RightArc.cost
elif move == BREAK:
t.is_valid = Break.is_valid
t.do = Break.transition
t.get_cost = Break.cost
else:
raise Exception(move)
return t
cdef int initialize_state(self, StateC* st) nogil:
for i in range(st.length):
if st._sent[i].dep == 0:
st._sent[i].l_edge = i
st._sent[i].r_edge = i
st._sent[i].head = 0
st._sent[i].dep = 0
st._sent[i].l_kids = 0
st._sent[i].r_kids = 0
st.fast_forward()
cdef int finalize_state(self, StateC* st) nogil:
cdef int i
for i in range(st.length):
if st._sent[i].head == 0:
st._sent[i].dep = self.root_label
def finalize_doc(self, doc):
doc.is_parsed = True
2016-02-01 05:07:37 +03:00
cdef int set_valid(self, int* output, const StateC* st) nogil:
cdef bint[N_MOVES] is_valid
2017-05-30 21:37:24 +03:00
is_valid[SHIFT] = Shift.is_valid(st, 0)
is_valid[REDUCE] = Reduce.is_valid(st, 0)
is_valid[LEFT] = LeftArc.is_valid(st, 0)
is_valid[RIGHT] = RightArc.is_valid(st, 0)
is_valid[BREAK] = Break.is_valid(st, 0)
cdef int i
for i in range(self.n_moves):
output[i] = is_valid[self.c[i].move]
cdef int set_costs(self, int* is_valid, weight_t* costs,
StateClass stcls, GoldParse gold) except -1:
2017-05-30 21:37:24 +03:00
cdef int i, move
cdef attr_t label
cdef label_cost_func_t[N_MOVES] label_cost_funcs
cdef move_cost_func_t[N_MOVES] move_cost_funcs
cdef weight_t[N_MOVES] move_costs
for i in range(N_MOVES):
move_costs[i] = 9000
move_cost_funcs[SHIFT] = Shift.move_cost
move_cost_funcs[REDUCE] = Reduce.move_cost
move_cost_funcs[LEFT] = LeftArc.move_cost
move_cost_funcs[RIGHT] = RightArc.move_cost
move_cost_funcs[BREAK] = Break.move_cost
label_cost_funcs[SHIFT] = Shift.label_cost
label_cost_funcs[REDUCE] = Reduce.label_cost
label_cost_funcs[LEFT] = LeftArc.label_cost
label_cost_funcs[RIGHT] = RightArc.label_cost
label_cost_funcs[BREAK] = Break.label_cost
2017-05-28 15:06:40 +03:00
cdef attr_t* labels = gold.c.labels
cdef int* heads = gold.c.heads
n_gold = 0
for i in range(self.n_moves):
if self.c[i].is_valid(stcls.c, self.c[i].label):
is_valid[i] = True
move = self.c[i].move
label = self.c[i].label
if move_costs[move] == 9000:
move_costs[move] = move_cost_funcs[move](stcls, &gold.c)
costs[i] = move_costs[move] + label_cost_funcs[move](stcls, &gold.c, label)
n_gold += costs[i] <= 0
2015-06-10 07:56:35 +03:00
else:
is_valid[i] = False
costs[i] = 9000
if n_gold < 1:
# Check label set --- leading cause
label_set = set([self.strings[self.c[i].label] for i in range(self.n_moves)])
for label_str in gold.labels:
if label_str is not None and label_str not in label_set:
raise ValueError("Cannot get gold parser action: unknown label: %s" % label_str)
# Check projectivity --- other leading cause
Improve label management in parser and NER (#2108) This patch does a few smallish things that tighten up the training workflow a little, and allow memory use during training to be reduced by letting the GoldCorpus stream data properly. Previously, the parser and entity recognizer read and saved labels as lists, with extra labels noted separately. Lists were used becaue ordering is very important, to ensure that the label-to-class mapping is stable. We now manage labels as nested dictionaries, first keyed by the action, and then keyed by the label. Values are frequencies. The trick is, how do we save new labels? We need to make sure we iterate over these in the same order they're added. Otherwise, we'll get different class IDs, and the model's predictions won't make sense. To allow stable sorting, we map the new labels to negative values. If we have two new labels, they'll be noted as having "frequency" -1 and -2. The next new label will then have "frequency" -3. When we sort by (frequency, label), we then get a stable sort. Storing frequencies then allows us to make the next nice improvement. Previously we had to iterate over the whole training set, to pre-process it for the deprojectivisation. This led to storing the whole training set in memory. This was most of the required memory during training. To prevent this, we now store the frequencies as we stream in the data, and deprojectivize as we go. Once we've built the frequencies, we can then apply a frequency cut-off when we decide how many classes to make. Finally, to allow proper data streaming, we also have to have some way of shuffling the iterator. This is awkward if the training files have multiple documents in them. To solve this, the GoldCorpus class now writes the training data to disk in msgpack files, one per document. We can then shuffle the data by shuffling the paths. This is a squash merge, as I made a lot of very small commits. Individual commit messages below. * Simplify label management for TransitionSystem and its subclasses * Fix serialization for new label handling format in parser * Simplify and improve GoldCorpus class. Reduce memory use, write to temp dir * Set actions in transition system * Require thinc 6.11.1.dev4 * Fix error in parser init * Add unicode declaration * Fix unicode declaration * Update textcat test * Try to get model training on less memory * Print json loc for now * Try rapidjson to reduce memory use * Remove rapidjson requirement * Try rapidjson for reduced mem usage * Handle None heads when projectivising * Stream json docs * Fix train script * Handle projectivity in GoldParse * Fix projectivity handling * Add minibatch_by_words util from ud_train * Minibatch by number of words in spacy.cli.train * Move minibatch_by_words util to spacy.util * Fix label handling * More hacking at label management in parser * Fix encoding in msgpack serialization in GoldParse * Adjust batch sizes in parser training * Fix minibatch_by_words * Add merge_subtokens function to pipeline.pyx * Register merge_subtokens factory * Restore use of msgpack tmp directory * Use minibatch-by-words in train * Handle retokenization in scorer * Change back-off approach for missing labels. Use 'dep' label * Update NER for new label management * Set NER tags for over-segmented words * Fix label alignment in gold * Fix label back-off for infrequent labels * Fix int type in labels dict key * Fix int type in labels dict key * Update feature definition for 8 feature set * Update ud-train script for new label stuff * Fix json streamer * Print the line number if conll eval fails * Update children and sentence boundaries after deprojectivisation * Export set_children_from_heads from doc.pxd * Render parses during UD training * Remove print statement * Require thinc 6.11.1.dev6. Try adding wheel as install_requires * Set different dev version, to flush pip cache * Update thinc version * Update GoldCorpus docs * Remove print statements * Fix formatting and links [ci skip]
2018-03-19 04:58:08 +03:00
if nonproj.is_nonproj_tree(gold.heads):
raise ValueError(
2017-10-27 20:45:57 +03:00
"Could not find a gold-standard action to supervise the "
"dependency parser. Likely cause: the tree is "
"non-projective (i.e. it has crossing arcs -- see "
"spacy/syntax/nonproj.pyx for definitions). The ArcEager "
"transition system only supports projective trees. To "
"learn non-projective representations, transform the data "
"before training and after parsing. Either pass "
"make_projective=True to the GoldParse class, or use "
"spacy.syntax.nonproj.preprocess_training_data.")
else:
2017-08-26 03:50:55 +03:00
print(gold.orig_annot)
print(gold.words)
print(gold.heads)
print(gold.labels)
2017-08-26 03:50:55 +03:00
print(gold.sent_starts)
raise ValueError(
2017-10-27 20:45:57 +03:00
"Could not find a gold-standard action to supervise the"
"dependency parser. The GoldParse was projective. The "
"transition system has %d actions. State at failure: %s"
% (self.n_moves, stcls.print_state(gold.words)))
assert n_gold >= 1
def get_beam_annot(self, Beam beam):
2017-11-14 04:11:40 +03:00
length = (<StateC*>beam.at(0)).length
heads = [{} for _ in range(length)]
deps = [{} for _ in range(length)]
probs = beam.probs
for i in range(beam.size):
2017-11-14 04:11:40 +03:00
state = <StateC*>beam.at(i)
self.finalize_state(state)
if state.is_final():
prob = probs[i]
2017-11-14 04:11:40 +03:00
for j in range(state.length):
head = j + state._sent[j].head
dep = state._sent[j].dep
heads[j].setdefault(head, 0.0)
heads[j][head] += prob
deps[j].setdefault(dep, 0.0)
deps[j][dep] += prob
return heads, deps