def _train_classifier(self, data_class, encoder_name, classifier_name):
learn = text_classifier_learner(data_class, AWD_LSTM, drop_mult=0.5, pretrained=False)
try:
path = f'{classifier_name}'
learn.load(path)
self.log('Loaded pretrained classifier from {}'.format(str(path)))
# return learn
except FileNotFoundError:
self.log(f'Training classifier `{classifier_name}`')
learn.load_encoder(encoder_name)
lr = 1e-1/LR_DIV_FACTOR
learn.fit_one_cycle(1, lr, moms=(0.8, 0.7))
# TODO remove
# learn.save(classifier_name)
# return learn
learn.freeze_to(-2)
lr /= 2
learn.fit_one_cycle(1, slice(lr/(2.6**4), lr), moms=(0.8, 0.7))
learn.freeze_to(-3)
lr /= 2
learn.fit_one_cycle(1, slice(lr/(2.6**4), lr), moms=(0.8, 0.7))
learn.unfreeze()
lr /= 5
learn.fit_one_cycle(2, slice(lr/(2.6**4), lr), moms=(0.8, 0.7))
self.log(f"Saving classifier `{classifier_name}`")
learn.save(classifier_name)
return learn
def train(bs):
path = Path("./")
data_lm = load_data(path, 'data_lm.pkl', bs=bs)
print("data_lm loaded")
data_clas = (TextList.from_csv(path, 'classifier.csv', cols=["summary", "description"], vocab=data_lm.vocab)
.split_from_df(col=3)
.label_from_df(cols=0)
.databunch(bs=bs))
print("data_clas loaded")
data_clas.show_batch() # not sure how it will work
data_clas.save('data_clas.pkl')
learn = text_classifier_learner(data_clas, AWD_LSTM, drop_mult=0.6)
learn.load_encoder('fine_tuned_enc')
lr_estimate = 1.0e-2
learn.fit_one_cycle(1, lr_estimate, moms=(0.8,0.7))
learn.save('first')
losses_fig = learn.recorder.plot_losses(return_fig=True)
losses_fig.savefig("losses_001.jpg", dpi=600)
def _fit_class(self, df_train, df_val, data_lm):
n_data = min(len(df_train), len(df_val))
# Classifier model data
data_class = TextClasDataBunch.from_df(
path="",
train_df=df_train,
valid_df=df_val,
vocab=data_lm.train_ds.vocab,
bs=self.batch_size if self.batch_size < n_data else n_data // 2,
)
# train the learner object
class_learner = text_classifier_learner(
data_class, self.arch, drop_mult=self.dropout_lm
)
class_learner.load_encoder(self.path_lm.name)
class_learner.fit_one_cycle(1, self.lr_class)
class_learner.export(self.path_class)
def _create_classifier_learner(self, data_lm, df_train, df_val, tokenizer,
encoder_name):
data_class = ftext.TextClasDataBunch.from_df(
'',
vocab=data_lm.train_ds.vocab,
bs=32,
train_df=df_train,
valid_df=df_val,
text_cols=0,
label_cols=1,
tokenizer=tokenizer,
label_delim=' ')
config = ftext.awd_lstm_clas_config.copy()
config['n_hid'] = 1150
learn = ftext.text_classifier_learner(data_class,
ftext.AWD_LSTM,
config=config,
drop_mult=0.5)
learn.load_encoder(encoder_name)
return learn
import ujson as json
import spacy
import fastai
from pathlib import Path
from fastai.text import load_data, text_classifier_learner, AWD_LSTM
from typing import Tuple
bs=48
path=Path('/app/model/sentiment')
data_clas = load_data(path, 'data_clas.pkl', bs=bs)
learn = text_classifier_learner(data_clas, AWD_LSTM, drop_mult=0.5)
learn.load('third')
learn.model.eval()
nlp = spacy.load('en_core_web_md')
def message_to_sentences(message:str, minimum_characters:int=5) -> str:
text = message
sentences = []
for line in text.split('\n'):
sentences += line.split('.')
return [s for s in sentences if len(s) > minimum_characters]
def predict_sentiment(sentence:str) -> Tuple[str, float]:
categorical, class_id, scores = learn.predict(sentence)
score = round(scores[class_id].item(), 4)
return "negative" if class_id == 0 else "positive", score
def model(message):
sentence = json.loads(message)
doc = nlp(sentence)
entities = nlp(sentence).ents
SAMPLES_PER_CLASS = 12500
print('loading data')
texts = []
target = []
for class_index, classname in enumerate(CLASS_NAMES):
for n, line in enumerate(open(DATA_FOLDER+classname+'.txt')):
texts.append(preprocess_string(line,False))
target.append(class_index)
if n > SAMPLES_PER_CLASS:
break
df = DataFrame({'label':target,'text':texts})
df_train, df_val = train_test_split(df, stratify = df['label'], test_size = 0.4, random_state = 12)
data_lm = TextLMDataBunch.from_df(train_df = df_train, valid_df = df_val, path = "")
data_clas = TextClasDataBunch.from_df(path = "", train_df = df_train, valid_df = df_val, vocab=data_lm.train_ds.vocab, bs=32)
learn = language_model_learner(data_lm, pretrained_model=URLs.WT103, drop_mult=0.7)
learn.fit_one_cycle(1, 1e-2)
learn.save_encoder('ft_enc')
learn = text_classifier_learner(data_clas, drop_mult=0.7)
learn.load_encoder('ft_enc')
learn.fit_one_cycle(1, 1e-2)
lm_learner.recorder.plot(suggestion=True)
lm_learner.fit_one_cycle(1, lm_learner.recorder.min_grad_lr)
lm_learner.save_encoder(model)
text_clas = TextClasDataBunch.from_df(
train_df=train_df,
valid_df=valid_df,
vocab=text_lm.train_ds.vocab,
path="",
)
clf = text_classifier_learner(
text_clas,
arch=AWD_LSTM,
drop_mult=0.2,
)
clf.load_encoder(model)
clf.lr_find()
clf.recorder.plot(suggestion=True)
clf.fit_one_cycle(1, clf.recorder.min_grad_lr)
print(lm_learner.predict("green bell"))
print(text_clas.train_ds.y.c2i)
print(clf.predict("wrap avocado beef"))
def new_train_clas(data_dir,
lang='en',
cuda_id=0,
pretrain_name='wt103',
model_dir='models',
qrnn=False,
fine_tune=True,
max_vocab=30000,
bs=20,
bptt=70,
name='imdb-clas',
dataset='imdb',
ds_pct=1.0):
"""
:param data_dir: The path to the `data` directory
:param lang: the language unicode
:param cuda_id: The id of the GPU. Uses GPU 0 by default or no GPU when
run on CPU.
:param pretrain_name: name of the pretrained model
:param model_dir: The path to the directory where the pretrained model is saved
:param qrrn: Use a QRNN. Requires installing cupy.
:param fine_tune: Fine-tune the pretrained language model
:param max_vocab: The maximum size of the vocabulary.
:param bs: The batch size.
:param bptt: The back-propagation-through-time sequence length.
:param name: The name used for both the model and the vocabulary.
:param dataset: The dataset used for evaluation. Currently only IMDb and
XNLI are implemented. Assumes dataset is located in `data`
folder and that name of folder is the same as dataset name.
"""
results = {}
if not torch.cuda.is_available():
print('CUDA not available. Setting device=-1.')
cuda_id = -1
torch.cuda.set_device(cuda_id)
print(f'Dataset: {dataset}. Language: {lang}.')
assert dataset in DATASETS, f'Error: {dataset} processing is not implemented.'
assert (dataset == 'imdb' and lang == 'en') or not dataset == 'imdb',\
'Error: IMDb is only available in English.'
data_dir = Path(data_dir)
assert data_dir.name == 'data',\
f'Error: Name of data directory should be data, not {data_dir.name}.'
dataset_dir = data_dir / dataset
model_dir = Path(model_dir)
if qrnn:
print('Using QRNNs...')
model_name = 'qrnn' if qrnn else 'lstm'
lm_name = f'{model_name}_{pretrain_name}'
pretrained_fname = (lm_name, f'itos_{pretrain_name}')
ensure_paths_exists(data_dir, dataset_dir, model_dir,
model_dir / f"{pretrained_fname[0]}.pth",
model_dir / f"{pretrained_fname[1]}.pkl")
tmp_dir = dataset_dir / 'tmp'
tmp_dir.mkdir(exist_ok=True)
vocab_file = tmp_dir / f'vocab_{lang}.pkl'
if not (tmp_dir / f'{TRN}_{lang}_ids.npy').exists():
print('Reading the data...')
toks, lbls = read_clas_data(dataset_dir, dataset, lang)
# create the vocabulary
counter = Counter(word for example in toks[TRN] for word in example)
itos = [word for word, count in counter.most_common(n=max_vocab)]
itos.insert(0, PAD)
itos.insert(0, UNK)
vocab = Vocab(itos)
stoi = vocab.stoi
with open(vocab_file, 'wb') as f:
pickle.dump(vocab, f)
ids = {}
for split in [TRN, VAL, TST]:
ids[split] = np.array([([stoi.get(w, stoi[UNK]) for w in s])
for s in toks[split]])
np.save(tmp_dir / f'{split}_{lang}_ids.npy', ids[split])
np.save(tmp_dir / f'{split}_{lang}_lbl.npy', lbls[split])
else:
print('Loading the pickled data...')
ids, lbls = {}, {}
for split in [TRN, VAL, TST]:
ids[split] = np.load(tmp_dir / f'{split}_{lang}_ids.npy')
lbls[split] = np.load(tmp_dir / f'{split}_{lang}_lbl.npy')
with open(vocab_file, 'rb') as f:
vocab = pickle.load(f)
print(f'Train size: {len(ids[TRN])}. Valid size: {len(ids[VAL])}. '
f'Test size: {len(ids[TST])}.')
if ds_pct < 1.0:
print(f"Makeing the dataset smaller {ds_pct}")
for split in [TRN, VAL, TST]:
ids[split] = ids[split][:int(len(ids[split]) * ds_pct)]
data_lm = TextLMDataBunch.from_ids(path=tmp_dir,
vocab=vocab,
train_ids=ids[TRN],
valid_ids=ids[VAL],
bs=bs,
bptt=bptt)
# TODO TextClasDataBunch allows tst_ids as input, but not tst_lbls?
data_clas = TextClasDataBunch.from_ids(path=tmp_dir,
vocab=vocab,
train_ids=ids[TRN],
valid_ids=ids[VAL],
train_lbls=lbls[TRN],
valid_lbls=lbls[VAL],
bs=bs)
if qrnn:
emb_sz, nh, nl = 400, 1550, 3
else:
emb_sz, nh, nl = 400, 1150, 3
learn = language_model_learner(data_lm,
bptt=bptt,
emb_sz=emb_sz,
nh=nh,
nl=nl,
qrnn=qrnn,
pad_token=PAD_TOKEN_ID,
pretrained_fnames=pretrained_fname,
path=model_dir.parent,
model_dir=model_dir.name)
lm_enc_finetuned = f"{lm_name}_{dataset}_enc"
if fine_tune and not (model_dir / f"lm_enc_finetuned.pth").exists():
print('Fine-tuning the language model...')
learn.unfreeze()
learn.fit(2, slice(1e-4, 1e-2))
# save encoder
learn.save_encoder(lm_enc_finetuned)
print("Starting classifier training")
learn = text_classifier_learner(data_clas,
bptt=bptt,
pad_token=PAD_TOKEN_ID,
path=model_dir.parent,
model_dir=model_dir.name,
qrnn=qrnn,
emb_sz=emb_sz,
nh=nh,
nl=nl)
learn.load_encoder(lm_enc_finetuned)
learn.fit_one_cycle(1, 2e-2, moms=(0.8, 0.7), wd=1e-7)
learn.freeze_to(-2)
learn.fit_one_cycle(1,
slice(1e-2 / (2.6**4), 1e-2),
moms=(0.8, 0.7),
wd=1e-7)
learn.freeze_to(-3)
learn.fit_one_cycle(1,
slice(5e-3 / (2.6**4), 5e-3),
moms=(0.8, 0.7),
wd=1e-7)
learn.unfreeze()
learn.fit_one_cycle(2,
slice(1e-3 / (2.6**4), 1e-3),
moms=(0.8, 0.7),
wd=1e-7)
results['accuracy'] = learn.validate()[1]
print(f"Saving models at {learn.path / learn.model_dir}")
learn.save(f'{model_name}_{name}')
return results
