Commit Graph

5059 Commits

Author SHA1 Message Date
Matthew Honnibal
5cac951a16 Move new parser to nn_parser.pyx, and restore old parser, to make tests pass. 2017-05-14 00:55:01 +02:00
Matthew Honnibal
f8c02b4341 Remove cupy imports from parser, so it can work on CPU 2017-05-14 00:37:53 +02:00
Matthew Honnibal
613ba79e2e Fiddle with sizings for parser 2017-05-13 17:20:23 -05:00
Matthew Honnibal
e6d71e1778 Small fixes to parser 2017-05-13 17:19:04 -05:00
Matthew Honnibal
188c0f6949 Clean up unused import 2017-05-13 17:18:27 -05:00
Matthew Honnibal
f85c8464f7 Draft support of regression loss in parser 2017-05-13 17:17:27 -05:00
Matthew Honnibal
827b5af697 Update draft of parser neural network model
Model is good, but code is messy. Currently requires Chainer, which may cause the build to fail on machines without a GPU.

Outline of the model:

We first predict context-sensitive vectors for each word in the input:

(embed_lower | embed_prefix | embed_suffix | embed_shape)
>> Maxout(token_width)
>> convolution ** 4

This convolutional layer is shared between the tagger and the parser. This prevents the parser from needing tag features.
To boost the representation, we make a "super tag" with POS, morphology and dependency label. The tagger predicts this
by adding a softmax layer onto the convolutional layer --- so, we're teaching the convolutional layer to give us a
representation that's one affine transform from this informative lexical information. This is obviously good for the
parser (which backprops to the convolutions too).

The parser model makes a state vector by concatenating the vector representations for its context tokens. Current
results suggest few context tokens works well. Maybe this is a bug.

The current context tokens:

* S0, S1, S2: Top three words on the stack
* B0, B1: First two words of the buffer
* S0L1, S0L2: Leftmost and second leftmost children of S0
* S0R1, S0R2: Rightmost and second rightmost children of S0
* S1L1, S1L2, S1R2, S1R, B0L1, B0L2: Likewise for S1 and B0

This makes the state vector quite long: 13*T, where T is the token vector width (128 is working well). Fortunately,
there's a way to structure the computation to save some expense (and make it more GPU friendly).

The parser typically visits 2*N states for a sentence of length N (although it may visit more, if it back-tracks
with a non-monotonic transition). A naive implementation would require 2*N (B, 13*T) @ (13*T, H) matrix multiplications
for a batch of size B. We can instead perform one (B*N, T) @ (T, 13*H) multiplication, to pre-compute the hidden
weights for each positional feature wrt the words in the batch. (Note that our token vectors come from the CNN
-- so we can't play this trick over the vocabulary. That's how Stanford's NN parser works --- and why its model
is so big.)

This pre-computation strategy allows a nice compromise between GPU-friendliness and implementation simplicity.
The CNN and the wide lower layer are computed on the GPU, and then the precomputed hidden weights are moved
to the CPU, before we start the transition-based parsing process. This makes a lot of things much easier.
We don't have to worry about variable-length batch sizes, and we don't have to implement the dynamic oracle
in CUDA to train.

Currently the parser's loss function is multilabel log loss, as the dynamic oracle allows multiple states to
be 0 cost. This is defined as:

(exp(score) / Z) - (exp(score) / gZ)

Where gZ is the sum of the scores assigned to gold classes. I'm very interested in regressing on the cost directly,
but so far this isn't working well.

Machinery is in place for beam-search, which has been working well for the linear model. Beam search should benefit
greatly from the pre-computation trick.
2017-05-12 16:09:15 -05:00
Matthew Honnibal
b44f7e259c Clean up unused parser code 2017-05-08 15:42:04 +02:00
Matthew Honnibal
17efb1c001 Change width 2017-05-08 08:40:13 -05:00
Matthew Honnibal
5dffb85184 Don't use gpu 2017-05-08 08:39:59 -05:00
Matthew Honnibal
bef89ef23d Mergery 2017-05-08 08:29:36 -05:00
Matthew Honnibal
245372973d Don't use tagger to predict tags 2017-05-08 07:55:34 -05:00
Matthew Honnibal
50ddc9fc45 Fix infinite loop bug 2017-05-08 07:54:26 -05:00
Matthew Honnibal
94e86ae00a Predict tags with encoder 2017-05-08 07:53:45 -05:00
Matthew Honnibal
56073a11ef Don't use tags when calculating token vectors 2017-05-08 07:52:24 -05:00
Matthew Honnibal
7a33f1e2b7 Add dep to supertag. 2017-05-08 07:50:01 -05:00
Matthew Honnibal
66252f3e71 Change vector width 2017-05-08 14:47:11 +02:00
Matthew Honnibal
a66a4a4d0f Replace einsums 2017-05-08 14:46:50 +02:00
Matthew Honnibal
8d2eab74da Use PretrainableMaxouts 2017-05-08 14:24:55 +02:00
Matthew Honnibal
807cb2e370 Add PretrainableMaxouts 2017-05-08 14:24:43 +02:00
Matthew Honnibal
2e2268a442 Precomputable hidden now working 2017-05-08 11:36:37 +02:00
Matthew Honnibal
10682d35ab Get pre-computed version working 2017-05-08 00:38:35 +02:00
Matthew Honnibal
35458987e8 Checkpoint -- nearly finished reimpl 2017-05-07 23:05:01 +02:00
Matthew Honnibal
4441866f55 Checkpoint -- nearly finished reimpl 2017-05-07 22:47:06 +02:00
Matthew Honnibal
6782eedf9b Tmp GPU code 2017-05-07 11:04:24 -05:00
Matthew Honnibal
e420e5a809 Tmp 2017-05-07 07:31:09 -05:00
Matthew Honnibal
12039e80ca Switch to single matmul for state layer 2017-05-07 14:26:34 +02:00
Matthew Honnibal
700979fb3c CPU/GPU compat 2017-05-07 04:01:11 +02:00
Matthew Honnibal
f99f5b75dc working residual net 2017-05-07 03:57:26 +02:00
Matthew Honnibal
bdf2dba9fb WIP on refactor, with hidde pre-computing 2017-05-07 02:02:43 +02:00
Matthew Honnibal
b439e04f8d Learning smoothly 2017-05-06 20:38:12 +02:00
Matthew Honnibal
08bee76790 Learns things 2017-05-06 18:24:38 +02:00
Matthew Honnibal
04ae1c01f1 Learns things 2017-05-06 18:21:02 +02:00
Matthew Honnibal
bcf4cd0a5f Learns things 2017-05-06 17:37:36 +02:00
Matthew Honnibal
8e48b58cd6 Gradients look correct 2017-05-06 16:47:15 +02:00
Matthew Honnibal
7e04260d38 Data running through, likely errors in model 2017-05-06 14:22:20 +02:00
Matthew Honnibal
fa7c1990b6 Restore tok2vec function 2017-05-05 20:12:03 +02:00
Matthew Honnibal
efe9630e1c Bug fixes 2017-05-05 20:09:50 +02:00
Matthew Honnibal
ef4fa594aa Draft of NN parser, to be tested 2017-05-05 19:20:39 +02:00
Matthew Honnibal
7d1df50aec Draft up Parser model 2017-05-04 13:31:40 +02:00
Matthew Honnibal
ccaf26206b Pseudocode for parser 2017-05-04 12:17:59 +02:00
Ines Montani
6e1fad92a1 Update CONTRIBUTORS.md 2017-05-03 10:01:40 +02:00
ines
e2380d8789 Update README.rst 2017-05-03 10:00:04 +02:00
ines
f9384b0fbd Update alpha languages and add aside for tokenizer dependencies 2017-05-03 09:58:31 +02:00
Ines Montani
f0d7a87e18 Merge pull request #1035 from uetchy/japanese-support
Japanese support
2017-05-03 09:44:54 +02:00
Ines Montani
3ea23a3f4d Fix formatting 2017-05-03 09:44:38 +02:00
Ines Montani
d730eb0c0d Raise custom ImportError if importing janome fails 2017-05-03 09:43:29 +02:00
Ines Montani
949ad6594b Add newline 2017-05-03 09:38:43 +02:00
Ines Montani
d12ca587ea Add newline 2017-05-03 09:38:29 +02:00
Ines Montani
8676cd0135 Add newline 2017-05-03 09:38:07 +02:00