spaCy/spacy/tests/test_models.py

188 lines
5.9 KiB
Python
Raw Normal View History

from typing import List
import pytest
from thinc.api import fix_random_seed, Adam, set_dropout_rate
from numpy.testing import assert_array_equal
import numpy
2020-07-28 20:43:19 +00:00
from spacy.ml.models import build_Tok2Vec_model, MultiHashEmbed, MaxoutWindowEncoder
from spacy.ml.models import build_text_classifier, build_simple_cnn_text_classifier
from spacy.lang.en import English
from spacy.lang.en.examples import sentences as EN_SENTENCES
2020-07-28 20:43:19 +00:00
def get_textcat_kwargs():
return {
"width": 64,
"embed_size": 2000,
"pretrained_vectors": None,
"exclusive_classes": False,
"ngram_size": 1,
"window_size": 1,
"conv_depth": 2,
"dropout": None,
"nO": 7,
}
2020-08-05 14:00:59 +00:00
2020-07-28 20:43:19 +00:00
def get_textcat_cnn_kwargs():
return {
"tok2vec": test_tok2vec(),
"exclusive_classes": False,
"nO": 13,
}
2020-08-05 14:00:59 +00:00
def get_all_params(model):
params = []
for node in model.walk():
for name in node.param_names:
params.append(node.get_param(name).ravel())
return node.ops.xp.concatenate(params)
def get_docs():
nlp = English()
return list(nlp.pipe(EN_SENTENCES + [" ".join(EN_SENTENCES)]))
def get_gradient(model, Y):
if isinstance(Y, model.ops.xp.ndarray):
dY = model.ops.alloc(Y.shape, dtype=Y.dtype)
dY += model.ops.xp.random.uniform(-1.0, 1.0, Y.shape)
return dY
elif isinstance(Y, List):
return [get_gradient(model, y) for y in Y]
else:
raise ValueError(f"Could not get gradient for type {type(Y)}")
2020-07-28 20:43:19 +00:00
def get_tok2vec_kwargs():
# This actually creates models, so seems best to put it in a function.
return {
"embed": MultiHashEmbed(
width=32,
2020-10-05 17:57:45 +00:00
rows=[500, 500, 500],
attrs=["NORM", "PREFIX", "SHAPE"],
2020-10-05 19:58:18 +00:00
include_static_vectors=False,
2020-07-28 20:43:19 +00:00
),
"encode": MaxoutWindowEncoder(
2020-09-29 19:39:28 +00:00
width=32, depth=2, maxout_pieces=2, window_size=1
2020-08-05 14:00:59 +00:00
),
2020-07-28 20:43:19 +00:00
}
Refactor pipeline components, config and language data (#5759) * Update with WIP * Update with WIP * Update with pipeline serialization * Update types and pipe factories * Add deep merge, tidy up and add tests * Fix pipe creation from config * Don't validate default configs on load * Update spacy/language.py Co-authored-by: Ines Montani <ines@ines.io> * Adjust factory/component meta error * Clean up factory args and remove defaults * Add test for failing empty dict defaults * Update pipeline handling and methods * provide KB as registry function instead of as object * small change in test to make functionality more clear * update example script for EL configuration * Fix typo * Simplify test * Simplify test * splitting pipes.pyx into separate files * moving default configs to each component file * fix batch_size type * removing default values from component constructors where possible (TODO: test 4725) * skip instead of xfail * Add test for config -> nlp with multiple instances * pipeline.pipes -> pipeline.pipe * Tidy up, document, remove kwargs * small cleanup/generalization for Tok2VecListener * use DEFAULT_UPSTREAM field * revert to avoid circular imports * Fix tests * Replace deprecated arg * Make model dirs require config * fix pickling of keyword-only arguments in constructor * WIP: clean up and integrate full config * Add helper to handle function args more reliably Now also includes keyword-only args * Fix config composition and serialization * Improve config debugging and add visual diff * Remove unused defaults and fix type * Remove pipeline and factories from meta * Update spacy/default_config.cfg Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com> * Update spacy/default_config.cfg * small UX edits * avoid printing stack trace for debug CLI commands * Add support for language-specific factories * specify the section of the config which holds the model to debug * WIP: add Language.from_config * Update with language data refactor WIP * Auto-format * Add backwards-compat handling for Language.factories * Update morphologizer.pyx * Fix morphologizer * Update and simplify lemmatizers * Fix Japanese tests * Port over tagger changes * Fix Chinese and tests * Update to latest Thinc * WIP: xfail first Russian lemmatizer test * Fix component-specific overrides * fix nO for output layers in debug_model * Fix default value * Fix tests and don't pass objects in config * Fix deep merging * Fix lemma lookup data registry Only load the lookups if an entry is available in the registry (and if spacy-lookups-data is installed) * Add types * Add Vocab.from_config * Fix typo * Fix tests * Make config copying more elegant * Fix pipe analysis * Fix lemmatizers and is_base_form * WIP: move language defaults to config * Fix morphology type * Fix vocab * Remove comment * Update to latest Thinc * Add morph rules to config * Tidy up * Remove set_morphology option from tagger factory * Hack use_gpu * Move [pipeline] to top-level block and make [nlp.pipeline] list Allows separating component blocks from component order – otherwise, ordering the config would mean a changed component order, which is bad. Also allows initial config to define more components and not use all of them * Fix use_gpu and resume in CLI * Auto-format * Remove resume from config * Fix formatting and error * [pipeline] -> [components] * Fix types * Fix tagger test: requires set_morphology? Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com> Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com> Co-authored-by: Matthew Honnibal <honnibal+gh@gmail.com>
2020-07-22 11:42:59 +00:00
def test_tok2vec():
2020-07-28 20:43:19 +00:00
return build_Tok2Vec_model(**get_tok2vec_kwargs())
def test_multi_hash_embed():
embed = MultiHashEmbed(
width=32,
2020-10-05 17:57:45 +00:00
rows=[500, 500, 500],
attrs=["NORM", "PREFIX", "SHAPE"],
2020-10-05 19:58:18 +00:00
include_static_vectors=False,
)
hash_embeds = [node for node in embed.walk() if node.name == "hashembed"]
assert len(hash_embeds) == 3
# Check they look at different columns.
assert list(sorted(he.attrs["column"] for he in hash_embeds)) == [0, 1, 2]
# Check they use different seeds
assert len(set(he.attrs["seed"] for he in hash_embeds)) == 3
# Check they all have the same number of rows
assert [he.get_dim("nV") for he in hash_embeds] == [500, 500, 500]
# Now try with different row factors
embed = MultiHashEmbed(
width=32,
2020-10-05 17:57:45 +00:00
rows=[1000, 50, 250],
attrs=["NORM", "PREFIX", "SHAPE"],
2020-10-05 19:58:18 +00:00
include_static_vectors=False,
)
hash_embeds = [node for node in embed.walk() if node.name == "hashembed"]
assert [he.get_dim("nV") for he in hash_embeds] == [1000, 50, 250]
2020-10-05 19:58:18 +00:00
@pytest.mark.parametrize(
"seed,model_func,kwargs",
[
2020-07-28 20:43:19 +00:00
(0, build_Tok2Vec_model, get_tok2vec_kwargs()),
(0, build_text_classifier, get_textcat_kwargs()),
(0, build_simple_cnn_text_classifier, get_textcat_cnn_kwargs()),
],
)
def test_models_initialize_consistently(seed, model_func, kwargs):
fix_random_seed(seed)
model1 = model_func(**kwargs)
model1.initialize()
fix_random_seed(seed)
model2 = model_func(**kwargs)
model2.initialize()
params1 = get_all_params(model1)
params2 = get_all_params(model2)
assert_array_equal(params1, params2)
@pytest.mark.parametrize(
"seed,model_func,kwargs,get_X",
[
2020-07-28 20:43:19 +00:00
(0, build_Tok2Vec_model, get_tok2vec_kwargs(), get_docs),
(0, build_text_classifier, get_textcat_kwargs(), get_docs),
(0, build_simple_cnn_text_classifier, get_textcat_cnn_kwargs(), get_docs),
],
)
def test_models_predict_consistently(seed, model_func, kwargs, get_X):
fix_random_seed(seed)
model1 = model_func(**kwargs).initialize()
Y1 = model1.predict(get_X())
fix_random_seed(seed)
model2 = model_func(**kwargs).initialize()
Y2 = model2.predict(get_X())
if model1.has_ref("tok2vec"):
tok2vec1 = model1.get_ref("tok2vec").predict(get_X())
tok2vec2 = model2.get_ref("tok2vec").predict(get_X())
for i in range(len(tok2vec1)):
for j in range(len(tok2vec1[i])):
Refactor pipeline components, config and language data (#5759) * Update with WIP * Update with WIP * Update with pipeline serialization * Update types and pipe factories * Add deep merge, tidy up and add tests * Fix pipe creation from config * Don't validate default configs on load * Update spacy/language.py Co-authored-by: Ines Montani <ines@ines.io> * Adjust factory/component meta error * Clean up factory args and remove defaults * Add test for failing empty dict defaults * Update pipeline handling and methods * provide KB as registry function instead of as object * small change in test to make functionality more clear * update example script for EL configuration * Fix typo * Simplify test * Simplify test * splitting pipes.pyx into separate files * moving default configs to each component file * fix batch_size type * removing default values from component constructors where possible (TODO: test 4725) * skip instead of xfail * Add test for config -> nlp with multiple instances * pipeline.pipes -> pipeline.pipe * Tidy up, document, remove kwargs * small cleanup/generalization for Tok2VecListener * use DEFAULT_UPSTREAM field * revert to avoid circular imports * Fix tests * Replace deprecated arg * Make model dirs require config * fix pickling of keyword-only arguments in constructor * WIP: clean up and integrate full config * Add helper to handle function args more reliably Now also includes keyword-only args * Fix config composition and serialization * Improve config debugging and add visual diff * Remove unused defaults and fix type * Remove pipeline and factories from meta * Update spacy/default_config.cfg Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com> * Update spacy/default_config.cfg * small UX edits * avoid printing stack trace for debug CLI commands * Add support for language-specific factories * specify the section of the config which holds the model to debug * WIP: add Language.from_config * Update with language data refactor WIP * Auto-format * Add backwards-compat handling for Language.factories * Update morphologizer.pyx * Fix morphologizer * Update and simplify lemmatizers * Fix Japanese tests * Port over tagger changes * Fix Chinese and tests * Update to latest Thinc * WIP: xfail first Russian lemmatizer test * Fix component-specific overrides * fix nO for output layers in debug_model * Fix default value * Fix tests and don't pass objects in config * Fix deep merging * Fix lemma lookup data registry Only load the lookups if an entry is available in the registry (and if spacy-lookups-data is installed) * Add types * Add Vocab.from_config * Fix typo * Fix tests * Make config copying more elegant * Fix pipe analysis * Fix lemmatizers and is_base_form * WIP: move language defaults to config * Fix morphology type * Fix vocab * Remove comment * Update to latest Thinc * Add morph rules to config * Tidy up * Remove set_morphology option from tagger factory * Hack use_gpu * Move [pipeline] to top-level block and make [nlp.pipeline] list Allows separating component blocks from component order – otherwise, ordering the config would mean a changed component order, which is bad. Also allows initial config to define more components and not use all of them * Fix use_gpu and resume in CLI * Auto-format * Remove resume from config * Fix formatting and error * [pipeline] -> [components] * Fix types * Fix tagger test: requires set_morphology? Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com> Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com> Co-authored-by: Matthew Honnibal <honnibal+gh@gmail.com>
2020-07-22 11:42:59 +00:00
assert_array_equal(
numpy.asarray(tok2vec1[i][j]), numpy.asarray(tok2vec2[i][j])
)
if isinstance(Y1, numpy.ndarray):
assert_array_equal(Y1, Y2)
elif isinstance(Y1, List):
assert len(Y1) == len(Y2)
for y1, y2 in zip(Y1, Y2):
assert_array_equal(y1, y2)
else:
raise ValueError(f"Could not compare type {type(Y1)}")
@pytest.mark.parametrize(
"seed,dropout,model_func,kwargs,get_X",
[
2020-07-28 20:43:19 +00:00
(0, 0.2, build_Tok2Vec_model, get_tok2vec_kwargs(), get_docs),
(0, 0.2, build_text_classifier, get_textcat_kwargs(), get_docs),
(0, 0.2, build_simple_cnn_text_classifier, get_textcat_cnn_kwargs(), get_docs),
],
)
def test_models_update_consistently(seed, dropout, model_func, kwargs, get_X):
def get_updated_model():
fix_random_seed(seed)
optimizer = Adam(0.001)
model = model_func(**kwargs).initialize()
initial_params = get_all_params(model)
set_dropout_rate(model, dropout)
for _ in range(5):
Y, get_dX = model.begin_update(get_X())
dY = get_gradient(model, Y)
Refactor pipeline components, config and language data (#5759) * Update with WIP * Update with WIP * Update with pipeline serialization * Update types and pipe factories * Add deep merge, tidy up and add tests * Fix pipe creation from config * Don't validate default configs on load * Update spacy/language.py Co-authored-by: Ines Montani <ines@ines.io> * Adjust factory/component meta error * Clean up factory args and remove defaults * Add test for failing empty dict defaults * Update pipeline handling and methods * provide KB as registry function instead of as object * small change in test to make functionality more clear * update example script for EL configuration * Fix typo * Simplify test * Simplify test * splitting pipes.pyx into separate files * moving default configs to each component file * fix batch_size type * removing default values from component constructors where possible (TODO: test 4725) * skip instead of xfail * Add test for config -> nlp with multiple instances * pipeline.pipes -> pipeline.pipe * Tidy up, document, remove kwargs * small cleanup/generalization for Tok2VecListener * use DEFAULT_UPSTREAM field * revert to avoid circular imports * Fix tests * Replace deprecated arg * Make model dirs require config * fix pickling of keyword-only arguments in constructor * WIP: clean up and integrate full config * Add helper to handle function args more reliably Now also includes keyword-only args * Fix config composition and serialization * Improve config debugging and add visual diff * Remove unused defaults and fix type * Remove pipeline and factories from meta * Update spacy/default_config.cfg Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com> * Update spacy/default_config.cfg * small UX edits * avoid printing stack trace for debug CLI commands * Add support for language-specific factories * specify the section of the config which holds the model to debug * WIP: add Language.from_config * Update with language data refactor WIP * Auto-format * Add backwards-compat handling for Language.factories * Update morphologizer.pyx * Fix morphologizer * Update and simplify lemmatizers * Fix Japanese tests * Port over tagger changes * Fix Chinese and tests * Update to latest Thinc * WIP: xfail first Russian lemmatizer test * Fix component-specific overrides * fix nO for output layers in debug_model * Fix default value * Fix tests and don't pass objects in config * Fix deep merging * Fix lemma lookup data registry Only load the lookups if an entry is available in the registry (and if spacy-lookups-data is installed) * Add types * Add Vocab.from_config * Fix typo * Fix tests * Make config copying more elegant * Fix pipe analysis * Fix lemmatizers and is_base_form * WIP: move language defaults to config * Fix morphology type * Fix vocab * Remove comment * Update to latest Thinc * Add morph rules to config * Tidy up * Remove set_morphology option from tagger factory * Hack use_gpu * Move [pipeline] to top-level block and make [nlp.pipeline] list Allows separating component blocks from component order – otherwise, ordering the config would mean a changed component order, which is bad. Also allows initial config to define more components and not use all of them * Fix use_gpu and resume in CLI * Auto-format * Remove resume from config * Fix formatting and error * [pipeline] -> [components] * Fix types * Fix tagger test: requires set_morphology? Co-authored-by: Sofie Van Landeghem <svlandeg@users.noreply.github.com> Co-authored-by: svlandeg <sofie.vanlandeghem@gmail.com> Co-authored-by: Matthew Honnibal <honnibal+gh@gmail.com>
2020-07-22 11:42:59 +00:00
get_dX(dY)
model.finish_update(optimizer)
updated_params = get_all_params(model)
with pytest.raises(AssertionError):
assert_array_equal(initial_params, updated_params)
return model
model1 = get_updated_model()
model2 = get_updated_model()
assert_array_equal(get_all_params(model1), get_all_params(model2))