spaCy/spacy/cli/train.py

130 lines
4.8 KiB
Python

# coding: utf8
from __future__ import unicode_literals, division, print_function
import json
from collections import defaultdict
import cytoolz
from pathlib import Path
import dill
import tqdm
from thinc.neural.optimizers import linear_decay
from ..tokens.doc import Doc
from ..scorer import Scorer
from ..gold import GoldParse, merge_sents
from ..gold import GoldCorpus
from ..util import prints
from .. import util
from .. import displacy
def train(lang_id, output_dir, train_data, dev_data, n_iter, n_sents,
use_gpu, no_tagger, no_parser, no_entities):
output_path = util.ensure_path(output_dir)
train_path = util.ensure_path(train_data)
dev_path = util.ensure_path(dev_data)
if not output_path.exists():
prints(output_path, title="Output directory not found", exits=True)
if not train_path.exists():
prints(train_path, title="Training data not found", exits=True)
if dev_path and not dev_path.exists():
prints(dev_path, title="Development data not found", exits=True)
lang_class = util.get_lang_class(lang_id)
pipeline = ['token_vectors', 'tags', 'dependencies', 'entities']
if no_tagger and 'tags' in pipeline: pipeline.remove('tags')
if no_parser and 'dependencies' in pipeline: pipeline.remove('dependencies')
if no_entities and 'entities' in pipeline: pipeline.remove('entities')
nlp = lang_class(pipeline=pipeline)
corpus = GoldCorpus(train_path, dev_path)
dropout = util.env_opt('dropout', 0.0)
dropout_decay = util.env_opt('dropout_decay', 0.0)
optimizer = nlp.begin_training(lambda: corpus.train_tuples, use_gpu=use_gpu)
n_train_docs = corpus.count_train()
batch_size = float(util.env_opt('min_batch_size', 4))
max_batch_size = util.env_opt('max_batch_size', 64)
batch_accel = util.env_opt('batch_accel', 1.001)
print("Itn.\tDep. Loss\tUAS\tNER F.\tTag %\tToken %")
for i in range(n_iter):
with tqdm.tqdm(total=n_train_docs) as pbar:
train_docs = corpus.train_docs(nlp, shuffle=i, projectivize=True)
idx = 0
while idx < n_train_docs:
batch = list(cytoolz.take(int(batch_size), train_docs))
if not batch:
break
docs, golds = zip(*batch)
nlp.update(docs, golds, drop=dropout, sgd=optimizer)
pbar.update(len(docs))
idx += len(docs)
batch_size *= batch_accel
batch_size = min(int(batch_size), max_batch_size)
dropout = linear_decay(dropout, dropout_decay, i*n_train_docs+idx)
with nlp.use_params(optimizer.averages):
scorer = nlp.evaluate(corpus.dev_docs(nlp))
print_progress(i, {}, scorer.scores)
with (output_path / 'model.bin').open('wb') as file_:
with nlp.use_params(optimizer.averages):
dill.dump(nlp, file_, -1)
def _render_parses(i, to_render):
to_render[0].user_data['title'] = "Batch %d" % i
with Path('/tmp/entities.html').open('w') as file_:
html = displacy.render(to_render[:5], style='ent', page=True)
file_.write(html)
with Path('/tmp/parses.html').open('w') as file_:
html = displacy.render(to_render[:5], style='dep', page=True)
file_.write(html)
def evaluate(Language, gold_tuples, path):
with (path / 'model.bin').open('rb') as file_:
nlp = dill.load(file_)
# TODO:
# 1. This code is duplicate with spacy.train.Trainer.evaluate
# 2. There's currently a semantic difference between pipe and
# not pipe! It matters whether we batch the inputs. Must fix!
all_docs = []
all_golds = []
for raw_text, paragraph_tuples in dev_sents:
if gold_preproc:
raw_text = None
else:
paragraph_tuples = merge_sents(paragraph_tuples)
docs = self.make_docs(raw_text, paragraph_tuples)
golds = self.make_golds(docs, paragraph_tuples)
all_docs.extend(docs)
all_golds.extend(golds)
scorer = Scorer()
for doc, gold in zip(self.nlp.pipe(all_docs), all_golds):
scorer.score(doc, gold)
return scorer
def print_progress(itn, losses, dev_scores):
# TODO: Fix!
scores = {}
for col in ['dep_loss', 'tag_loss', 'uas', 'tags_acc', 'token_acc', 'ents_f']:
scores[col] = 0.0
scores.update(losses)
scores.update(dev_scores)
tpl = '{:d}\t{dep_loss:.3f}\t{tag_loss:.3f}\t{uas:.3f}\t{ents_f:.3f}\t{tags_acc:.3f}\t{token_acc:.3f}'
print(tpl.format(itn, **scores))
def print_results(scorer):
results = {
'TOK': '%.2f' % scorer.token_acc,
'POS': '%.2f' % scorer.tags_acc,
'UAS': '%.2f' % scorer.uas,
'LAS': '%.2f' % scorer.las,
'NER P': '%.2f' % scorer.ents_p,
'NER R': '%.2f' % scorer.ents_r,
'NER F': '%.2f' % scorer.ents_f}
util.print_table(results, title="Results")