def test_overfitting(self):
layer_dim = 16
batcher = SrcBatcher(batch_size=10, break_ties_randomly=False)
train_args = {}
train_args['src_file'] = "examples/data/head.ja"
train_args['trg_file'] = "examples/data/head.en"
train_args['loss_calculator'] = MLELoss()
train_args['model'] = DefaultTranslator(
src_reader=PlainTextReader(vocab=Vocab(
vocab_file="examples/data/head.ja.vocab")),
trg_reader=PlainTextReader(vocab=Vocab(
vocab_file="examples/data/head.en.vocab")),
src_embedder=SimpleWordEmbedder(vocab_size=100, emb_dim=layer_dim),
encoder=BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim),
attender=MlpAttender(input_dim=layer_dim,
state_dim=layer_dim,
hidden_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(vocab_size=100, emb_dim=layer_dim),
decoder=AutoRegressiveDecoder(
input_dim=layer_dim,
trg_embed_dim=layer_dim,
rnn=UniLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_input_dim=layer_dim,
yaml_path="model.decoder.rnn"),
transform=NonLinear(input_dim=layer_dim * 2,
output_dim=layer_dim),
scorer=Softmax(input_dim=layer_dim, vocab_size=100),
bridge=CopyBridge(dec_dim=layer_dim, dec_layers=1)),
)
train_args['dev_tasks'] = [
LossEvalTask(model=train_args['model'],
src_file="examples/data/head.ja",
ref_file="examples/data/head.en",
batcher=batcher)
]
train_args['run_for_epochs'] = 1
train_args['trainer'] = AdamTrainer(alpha=0.1)
train_args['batcher'] = batcher
training_regimen = regimens.SimpleTrainingRegimen(**train_args)
for _ in range(50):
training_regimen.run_training(save_fct=lambda: None)
self.assertAlmostEqual(
0.0,
training_regimen.train_loss_tracker.epoch_loss.sum_factors() /
training_regimen.train_loss_tracker.epoch_words,
places=2)
def run(self):
seed = 13
random.seed(seed)
np.random.seed(seed)
EXP_DIR = os.path.dirname(__file__)
EXP = "annot"
model_file = f"{EXP_DIR}/results/{EXP}.mod"
log_file = f"{EXP_DIR}/results/{EXP}.log"
xnmt.tee.utils.dy.DynetParams().set_mem(
1024) #Doesnt work figure out how to set memory
xnmt.tee.set_out_file(log_file, exp_name=EXP)
ParamManager.init_param_col()
ParamManager.param_col.model_file = model_file
pre_runner = PreprocRunner(
tasks=[
PreprocTokenize(
in_files=
[ #f'{EXP_DIR}/conala-corpus/conala-trainnodev.snippet',
#f'{EXP_DIR}/conala-corpus/conala-trainnodev.intent',
#f'{EXP_DIR}/conala-corpus/conala-dev.intent',
#f'{EXP_DIR}/conala-corpus/conala-dev.snippet',
#f'{EXP_DIR}/conala-corpus/conala-test.intent',
#f'{EXP_DIR}/conala-corpus/conala-test.snippet',
f'{EXP_DIR}/conala-corpus/attack_code_train.txt',
f'{EXP_DIR}/conala-corpus/attack_text_train.txt',
f'{EXP_DIR}/conala-corpus/attack_code_test.txt',
f'{EXP_DIR}/conala-corpus/attack_text_test.txt'
#f'{EXP_DIR}/conala-corpus/all.code',
#f'{EXP_DIR}/conala-corpus/all.anno'
],
out_files=
[ #f'{EXP_DIR}/conala-corpus/conala-trainnodev.tmspm4000.snippet',
#f'{EXP_DIR}/conala-corpus/conala-trainnodev.tmspm4000.intent',
#f'{EXP_DIR}/conala-corpus/conala-dev.tmspm4000.intent',
#f'{EXP_DIR}/conala-corpus/conala-dev.tmspm4000.snippet',
#f'{EXP_DIR}/conala-corpus/conala-test.tmspm4000.intent',
#f'{EXP_DIR}/conala-corpus/conala-test.tmspm4000.snippet',
f'{EXP_DIR}/conala-corpus/attack-train.tmspm4000.snippet',
f'{EXP_DIR}/conala-corpus/attack-train.tmspm4000.intent',
f'{EXP_DIR}/conala-corpus/attack-test.tmspm4000.snippet',
f'{EXP_DIR}/conala-corpus/attack-test.tmspm4000.intent'
#f'{EXP_DIR}/conala-corpus/django.tmspm4000.snippet',
#f'{EXP_DIR}/conala-corpus/django.tmspm4000.intent'
],
specs=[{
'filenum':
'all',
'tokenizers': [
SentencepieceTokenizer(
hard_vocab_limit=False,
train_files=[
f'{EXP_DIR}/conala-corpus/attack_text_train.txt',
f'{EXP_DIR}/conala-corpus/attack_code_train.txt'
],
vocab_size=self.vocab_size,
model_type=self.model_type,
model_prefix=
'conala-corpus/attack-train.tmspm4000.spm')
]
}]),
PreprocVocab(
in_files=[
f'{EXP_DIR}/conala-corpus/attack-train.tmspm4000.intent',
f'{EXP_DIR}/conala-corpus/attack-train.tmspm4000.snippet'
],
out_files
=[
f'{EXP_DIR}/conala-corpus/attack-train.tmspm4000.intent.vocab',
f'{EXP_DIR}/conala-corpus/attack-train.tmspm4000.snippet.vocab'
],
specs=[{
'filenum':
'all',
'filters': [VocabFiltererFreq(min_freq=self.min_freq)]
}])
],
overwrite=False)
src_vocab = Vocab(
vocab_file=
f"{EXP_DIR}/conala-corpus/attack-train.tmspm4000.intent.vocab")
trg_vocab = Vocab(
vocab_file=
f"{EXP_DIR}/conala-corpus/attack-train.tmspm4000.snippet.vocab")
batcher = Batcher(batch_size=64)
inference = AutoRegressiveInference(search_strategy=BeamSearch(
len_norm=PolynomialNormalization(apply_during_search=True),
beam_size=5),
post_process='join-piece')
layer_dim = self.layer_dim
model = DefaultTranslator(
src_reader=PlainTextReader(vocab=src_vocab),
trg_reader=PlainTextReader(vocab=trg_vocab),
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim,
vocab=src_vocab),
encoder=BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim,
layers=self.layers),
attender=MlpAttender(hidden_dim=layer_dim,
state_dim=layer_dim,
input_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim,
vocab=trg_vocab),
decoder=AutoRegressiveDecoder(
input_dim=layer_dim,
rnn=UniLSTMSeqTransducer(
input_dim=layer_dim,
hidden_dim=layer_dim,
),
transform=AuxNonLinear(input_dim=layer_dim,
output_dim=layer_dim,
aux_input_dim=layer_dim),
scorer=Softmax(vocab_size=len(trg_vocab), input_dim=layer_dim),
trg_embed_dim=layer_dim,
input_feeding=False,
bridge=CopyBridge(dec_dim=layer_dim)),
inference=inference)
#decoder = AutoRegressiveDecoder(bridge=CopyBridge(),inference=inference))
train = SimpleTrainingRegimen(
name=f"{EXP}",
model=model,
batcher=WordSrcBatcher(avg_batch_size=64),
trainer=AdamTrainer(alpha=self.alpha),
patience=3,
lr_decay=0.5,
restart_trainer=True,
run_for_epochs=self.epochs,
src_file=f"{EXP_DIR}/conala-corpus/attack-train.tmspm4000.intent",
trg_file=f"{EXP_DIR}/conala-corpus/attack-train.tmspm4000.snippet",
dev_tasks=[
LossEvalTask(
src_file=
f"{EXP_DIR}/conala-corpus/attack-test.tmspm4000.intent",
ref_file=
f'{EXP_DIR}/conala-corpus/attack-test.tmspm4000.snippet',
model=model,
batcher=WordSrcBatcher(avg_batch_size=64)),
AccuracyEvalTask(
eval_metrics='bleu',
src_file=
f'{EXP_DIR}/conala-corpus/attack-test.tmspm4000.intent',
ref_file=f'{EXP_DIR}/conala-corpus/attack_text_test.txt',
hyp_file=f'results/{EXP}.dev.hyp',
model=model)
])
evaluate = [
AccuracyEvalTask(
eval_metrics="bleu",
#src_file=f"{EXP_DIR}/conala-corpus/conala-test.tmspm4000.intent",
src_file=
f"{EXP_DIR}/conala-corpus/attack-test.tmspm4000.intent",
#ref_file=f"{EXP_DIR}/conala-corpus/all.code",
#ref_file = f"{EXP_DIR}/conala-corpus/conala-test.snippet",
ref_file=f"{EXP_DIR}/conala-corpus/attack_text_test.txt",
hyp_file=f"results/{EXP}.test.hyp",
inference=inference,
model=model)
]
standard_experiment = Experiment(exp_global=ExpGlobal(
default_layer_dim=512,
dropout=0.3,
log_file=log_file,
model_file=model_file),
name="annot",
model=model,
train=train,
evaluate=evaluate)
# run experiment
standard_experiment(
save_fct=lambda: save_to_file(model_file, standard_experiment))
exit()
rnn=UniLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_input_dim=layer_dim,
yaml_path="decoder"),
transform=AuxNonLinear(input_dim=layer_dim,
output_dim=layer_dim,
aux_input_dim=layer_dim),
scorer=Softmax(vocab_size=len(trg_vocab), input_dim=layer_dim),
bridge=CopyBridge(dec_dim=layer_dim, dec_layers=1)),
inference=inference)
train = SimpleTrainingRegimen(
name=f"{EXP}",
model=model,
batcher=batcher,
trainer=AdamTrainer(alpha=0.001),
run_for_epochs=2,
src_file="examples/data/head.ja",
trg_file="examples/data/head.en",
dev_tasks=[
LossEvalTask(src_file="examples/data/head.ja",
ref_file="examples/data/head.en",
model=model,
batcher=batcher)
],
)
evaluate = [
AccuracyEvalTask(eval_metrics="bleu,wer",
src_file="examples/data/head.ja",
ref_file="examples/data/head.en",