def evaluate(model, data, params, split):
"""Evaluates a pretrained model on a dataset.
Inputs:
model (ATISModel): Model class.
data (ATISData): All of the data.
params (namespace): Parameters for the model.
"""
filename = split
if filename == 'dev':
split = data.dev_data
elif filename == 'train':
split = data.train_data
elif filename == 'test':
split = data.test_data
elif filename == 'valid':
split = data.valid_data
else:
raise ValueError("Split not recognized: " + str(params.evaluate_split))
if params.use_predicted_queries:
filename += "_use_predicted_queries"
else:
filename += "_use_gold_queries"
full_name = os.path.join(params.logdir, filename) + params.results_note
if params.interaction_level or params.use_predicted_queries:
examples = data.get_all_interactions(split)
if params.interaction_level:
valid_eval_results = evaluate_interaction_sample(
examples,
model,
name=full_name,
metrics=FINAL_EVAL_METRICS,
total_num=atis_data.num_utterances(split),
database_username=params.database_username,
database_password=params.database_password,
database_timeout=params.database_timeout,
use_predicted_queries=params.use_predicted_queries,
max_generation_length=params.eval_maximum_sql_length,
write_results=True,
use_gpu=True,
compute_metrics=params.compute_metrics)[0]
else:
valid_eval_results = evaluate_using_predicted_queries(
examples,
model,
name=full_name,
metrics=FINAL_EVAL_METRICS,
total_num=atis_data.num_utterances(split),
database_username=params.database_username,
database_password=params.database_password,
database_timeout=params.database_timeout)[0]
else:
examples = data.get_all_utterances(split)
valid_eval_results = evaluate_utterance_sample(
examples,
model,
name=full_name,
gold_forcing=False,
metrics=FINAL_EVAL_METRICS,
total_num=atis_data.num_utterances(split),
max_generation_length=params.eval_maximum_sql_length,
database_username=params.database_username,
database_password=params.database_password,
database_timeout=params.database_timeout,
write_results=True)[0]
for name, value in valid_eval_results.items():
print("valid gold-passing " + name.name + ":\t" + "%.2f" % value)
valid_token_accuracy = valid_eval_results[Metrics.TOKEN_ACCURACY]
string_accuracy = valid_eval_results[Metrics.STRING_ACCURACY]
print("token accuracy:\t" + str(valid_token_accuracy))
print("maximum string accuracy:\t" + str(string_accuracy))
def evaluate(model, data, params, last_save_file, split):
"""Evaluates a pretrained model on a dataset.
Inputs:
model (ATISModel): Model class.
data (ATISData): All of the data.
params (namespace): Parameters for the model.
last_save_file (str): Location where the model save file is.
"""
if last_save_file:
model.load(last_save_file)
else:
if not params.save_file:
raise ValueError(
"Must provide a save file name if not training first.")
model.load(params.save_file)
filename = split
if filename == 'dev':
split = data.dev_data
elif filename == 'train':
split = data.train_data
elif filename == 'test':
split = data.test_data
elif filename == 'valid':
split = data.valid_data
else:
raise ValueError("Split not recognized: " + str(params.evaluate_split))
if params.use_predicted_queries:
filename += "_use_predicted_queries"
else:
filename += "_use_gold_queries"
full_name = os.path.join(params.logdir, filename) + params.results_note
if params.interaction_level or params.use_predicted_queries:
examples = data.get_all_interactions(split)
if params.interaction_level:
evaluate_interaction_sample(
examples,
model,
name=full_name,
metrics=FINAL_EVAL_METRICS,
total_num=atis_data.num_utterances(split),
database_username=params.database_username,
database_password=params.database_password,
database_timeout=params.database_timeout,
use_predicted_queries=params.use_predicted_queries,
max_generation_length=params.eval_maximum_sql_length,
write_results=True,
use_gpu=True,
compute_metrics=params.compute_metrics)
else:
evaluate_using_predicted_queries(
examples,
model,
name=full_name,
metrics=FINAL_EVAL_METRICS,
total_num=atis_data.num_utterances(split),
database_username=params.database_username,
database_password=params.database_password,
database_timeout=params.database_timeout)
else:
examples = data.get_all_utterances(split)
evaluate_utterance_sample(
examples,
model,
name=full_name,
gold_forcing=False,
metrics=FINAL_EVAL_METRICS,
total_num=atis_data.num_utterances(split),
max_generation_length=params.eval_maximum_sql_length,
database_username=params.database_username,
database_password=params.database_password,
database_timeout=params.database_timeout,
write_results=True)
def evaluate(model, data, params, last_save_file):
if last_save_file:
model.load(last_save_file)
else:
if not params.save_file:
raise ValueError(
"Must provide a save file name if not training first.")
model.load(params.save_file)
split = None
if params.evaluate_split == 'dev':
split = data.dev_data
elif params.evaluate_split == 'train':
split = data.train_data
elif params.evaluate_split == 'test':
split = data.test_data
elif params.evaluate_split == 'valid':
split = data.valid_data
else:
raise ValueError("Split not recognized: " + str(params.evaluate_split))
filename = params.evaluate_split
if params.use_predicted_queries:
filename += "predicted"
else:
filename += "gold"
full_name = os.path.join(params.logdir, filename) + params.results_note
if params.interaction_level or params.use_predicted_queries:
examples = data.get_all_interactions(split)
if params.interaction_level:
results, _ = evaluate_interaction_sample(
examples,
model,
name=full_name,
metrics=FINAL_EVAL_METRICS,
total_num=atis_data.num_utterances(split),
database_username=params.database_username,
database_password=params.database_password,
database_timeout=params.database_timeout,
use_predicted_queries=params.use_predicted_queries,
max_generation_length=params.eval_maximum_sql_length,
write_results=True,
use_gpu=True)
else:
results, _ = evaluate_using_predicted_queries(
params,
examples,
model,
name=full_name,
metrics=FINAL_EVAL_METRICS,
total_num=atis_data.num_utterances(split),
max_generation_length=params.eval_maximum_sql_length,
database_username=params.database_username,
database_password=params.database_password,
database_timeout=params.database_timeout)
else:
examples = data.get_all_utterances(split)
results, _ = evaluate_utterance_sample(
examples,
model,
name=full_name,
gold_forcing=False,
metrics=FINAL_EVAL_METRICS,
total_num=atis_data.num_utterances(split),
max_generation_length=params.eval_maximum_sql_length,
database_username=params.database_username,
database_password=params.database_password,
database_timeout=params.database_timeout,
write_results=True)
