While developing v2.1, I ran a bunch of hyper-parameter search
experiments to find settings that performed well for spaCy's NER and
parser. I ended up changing the default Adam settings from beta1=0.9,
beta2=0.999, eps=1e-8 to beta1=0.8, beta2=0.8, eps=1e-5. This was giving
a small improvement in accuracy (like, 0.4%).
Months later, I run the models with Prodigy, which uses beam-search
decoding even when the model has been trained with a greedy objective.
The new models performed terribly...So, wtf? After a couple of days
debugging, I figured out that the new optimizer settings was causing the
model to converge to solutions where the top-scoring class often had
a score of like, -80. The variance on the weights had gone up
enormously. I guess I needed to update the L2 regularisation as well?
Anyway. Let's just revert the change --- if the optimizer is finding
such extreme solutions, that seems bad, and not nearly worth the small
improvement in accuracy.
Currently training a slate of models, to verify the accuracy change is minimal.
Once the training is complete, we can merge this.
<!--- Provide a general summary of your changes in the title. -->
## Description
<!--- Use this section to describe your changes. If your changes required
testing, include information about the testing environment and the tests you
ran. If your test fixes a bug reported in an issue, don't forget to include the
issue number. If your PR is still a work in progress, that's totally fine – just
include a note to let us know. -->
### Types of change
<!-- What type of change does your PR cover? Is it a bug fix, an enhancement
or new feature, or a change to the documentation? -->
## Checklist
<!--- Before you submit the PR, go over this checklist and make sure you can
tick off all the boxes. [] -> [x] -->
- [x] I have submitted the spaCy Contributor Agreement.
- [x] I ran the tests, and all new and existing tests passed.
- [x] My changes don't require a change to the documentation, or if they do, I've added all required information.
The new spacy pretrain command implemented BERT/ULMFit/etc-like transfer learning, using our Language Modelling with Approximate Outputs version of BERT's cloze task. Pretraining is convenient, but in some ways it's a bit of a strange solution. All we're doing is initialising the weights. At the same time, we're putting a lot of work into our optimisation so that it's less sensitive to initial conditions, and more likely to find good optima. I discuss this a bit in the pseudo-rehearsal blog post: https://explosion.ai/blog/pseudo-rehearsal-catastrophic-forgetting
Support semi-supervised learning in spacy train
One obvious way to improve these pretraining methods is to do multi-task learning, instead of just transfer learning. This has been shown to work very well: https://arxiv.org/pdf/1809.08370.pdf . This patch makes it easy to do this sort of thing.
Add a new argument to spacy train, --raw-text. This takes a jsonl file with unlabelled data that can be used in arbitrary ways to do semi-supervised learning.
Add a new method to the Language class and to pipeline components, .rehearse(). This is like .update(), but doesn't expect GoldParse objects. It takes a batch of Doc objects, and performs an update on some semi-supervised objective.
Move the BERT-LMAO objective out from spacy/cli/pretrain.py into spacy/_ml.py, so we can create a new pipeline component, ClozeMultitask. This can be specified as a parser or NER multitask in the spacy train command. Example usage:
python -m spacy train en ./tmp ~/data/en-core-web/train/nw.json ~/data/en-core-web/dev/nw.json --pipeline parser --raw-textt ~/data/unlabelled/reddit-100k.jsonl --vectors en_vectors_web_lg --parser-multitasks cloze
Implement rehearsal methods for pipeline components
The new --raw-text argument and nlp.rehearse() method also gives us a good place to implement the the idea in the pseudo-rehearsal blog post in the parser. This works as follows:
Add a new nlp.resume_training() method. This allocates copies of pre-trained models in the pipeline, setting things up for the rehearsal updates. It also returns an optimizer object. This also greatly reduces confusion around the nlp.begin_training() method, which randomises the weights, making it not suitable for adding new labels or otherwise fine-tuning a pre-trained model.
Implement rehearsal updates on the Parser class, making it available for the dependency parser and NER. During rehearsal, the initial model is used to supervise the model being trained. The current model is asked to match the predictions of the initial model on some data. This minimises catastrophic forgetting, by keeping the model's predictions close to the original. See the blog post for details.
Implement rehearsal updates for tagger
Implement rehearsal updates for text categoriz
Currently the TextCategorizer defaults to a fairly complicated model, designed partly around the active learning requirements of Prodigy. The model's a bit slow, and not very GPU-friendly.
This patch implements a straightforward CNN model that still performs pretty well. The replacement model also makes it easy to use the LMAO pretraining, since most of the parameters are in the CNN.
The replacement model has a flag to specify whether labels are mutually exclusive, which defaults to True. This has been a common problem with the text classifier. We'll also now be able to support adding labels to pretrained models again.
Resolves#2934, #2756, #1798, #1748.
<!--- Provide a general summary of your changes in the title. -->
## Description
- [x] Use [`black`](https://github.com/ambv/black) to auto-format all `.py` files.
- [x] Update flake8 config to exclude very large files (lemmatization tables etc.)
- [x] Update code to be compatible with flake8 rules
- [x] Fix various small bugs, inconsistencies and messy stuff in the language data
- [x] Update docs to explain new code style (`black`, `flake8`, when to use `# fmt: off` and `# fmt: on` and what `# noqa` means)
Once #2932 is merged, which auto-formats and tidies up the CLI, we'll be able to run `flake8 spacy` actually get meaningful results.
At the moment, the code style and linting isn't applied automatically, but I'm hoping that the new [GitHub Actions](https://github.com/features/actions) will let us auto-format pull requests and post comments with relevant linting information.
### Types of change
enhancement, code style
## Checklist
<!--- Before you submit the PR, go over this checklist and make sure you can
tick off all the boxes. [] -> [x] -->
- [x] I have submitted the spaCy Contributor Agreement.
- [x] I ran the tests, and all new and existing tests passed.
- [x] My changes don't require a change to the documentation, or if they do, I've added all required information.
These experiments were completed a few weeks ago, but I didn't make the PR, pending model release.
Token vector width: 128->96
Hidden width: 128->64
Embed size: 5000->2000
Dropout: 0.2->0.1
Updated optimizer defaults (unclear how important?)
This should improve speed, model size and load time, while keeping
similar or slightly better accuracy.
The tl;dr is we prefer to prevent over-fitting by reducing model size,
rather than using more dropout.
* Add spacy.errors module
* Update deprecation and user warnings
* Replace errors and asserts with new error message system
* Remove redundant asserts
* Fix whitespace
* Add messages for print/util.prints statements
* Fix typo
* Fix typos
* Move CLI messages to spacy.cli._messages
* Add decorator to display error code with message
An implementation like this is nice because it only modifies the string when it's retrieved from the containing class – so we don't have to worry about manipulating tracebacks etc.
* Remove unused link in spacy.about
* Update errors for invalid pipeline components
* Improve error for unknown factories
* Add displaCy warnings
* Update formatting consistency
* Move error message to spacy.errors
* Update errors and check if doc returned by component is None
This patch addresses #1660, which was caused by keying all pre-trained
vectors with the same ID when telling Thinc how to refer to them. This
meant that if multiple models were loaded that had pre-trained vectors,
errors or incorrect behaviour resulted.
The vectors class now includes a .name attribute, which defaults to:
{nlp.meta['lang']_nlp.meta['name']}.vectors
The vectors name is set in the cfg of the pipeline components under the
key pretrained_vectors. This replaces the previous cfg key
pretrained_dims.
In order to make existing models compatible with this change, we check
for the pretrained_dims key when loading models in from_disk and
from_bytes, and add the cfg key pretrained_vectors if we find it.
Matthew Honnibal