filename = dataset.dname + '_' + config_name + '.json'
print(f"############## {dataset.dname} ##############", flush=True)
tga = TGA(n_epochs=50,
lr=lr,
train_batch_size=bs,
w=w,
n_heads=nh,
weight_decay=l2,
device=f"cuda:{args.cuda}",
pretrained_vec=
'/home/mangaravite/Documents/pretrained_vectors/glove.6B.300d.txt',
verbose=True)
params = tga.get_params()
folds = []
for f, fold in enumerate(dataset.get_fold_instances(10)):
print(f"Fold {f}", flush=True)
tga.fit(fold.X_train, fold.y_train, fold.X_val, fold.y_val)
y_pred = tga.predict(fold.X_test)
correct = sum([p == l for (p, l) in zip(y_pred, fold.y_test)])
total = len(fold.X_test)
fold_out = {
'y_pred': list(map(int, list(y_pred))),
'y_test': list(map(int, list(fold.y_test))),
'epochs_acc': list(tga.checkpoints),
'model': tga.path_to_save
}
folds.append(fold_out)
print(f"END FOLD: ACC={(correct/total):.4f}")
params['folds'] = folds
with open(filename, 'w') as outfile:
d = Dataset(sys.argv[1])
path_result = path.join('result', d.dname)
if not path.exists(path_result):
mkdir(path_result)
space = {
"batch_size": hp.quniform("batch_size", 8, 128, 32),
"k": hp.quniform("k", 64, 512, 64),
"max_drop": hp.uniform("max_drop", 0.3, 0.9),
"mindf": hp.quniform("mindf", 2, 4, 1),
"stopwords": hp.choice("stopwords", ["nltk", "sklearn", None]),
}
for (i, fold) in enumerate(d.get_fold_instances(10, with_val=True)):
def hyperparameter_tuning_try(params):
try:
class_att = AttentionTFIDFClassifier(**params,
nepochs=25,
_verbose=False)
print(class_att)
class_att.fit(fold.X_train, fold.y_train, fold.X_val, fold.y_val)
return {
"loss": class_att._loss,
"status": STATUS_OK,
"model": class_att.to('cpu')
}
except:
return {"status": STATUS_FAIL}
