def main(config):
if config.mode == 'train':
train_data, valid_data = load_dataset(config.mode, config.random_seed)
# if use CNN model, pad sentences to let all the batch inputs has minimum length (filter_sizes[-1])
if config.model == 'cnn':
train_data = pad_sentence(train_data, config.filter_sizes[-1])
valid_data = pad_sentence(valid_data, config.filter_sizes[-1])
train_iter, valid_iter, pad_idx = make_iter(config.batch_size,
config.mode,
train_data=train_data,
valid_data=valid_data)
trainer = Trainer(config,
pad_idx,
train_iter=train_iter,
valid_iter=valid_iter)
trainer.train()
else:
test_data = load_dataset(config.mode, config.random_seed)
if config.model == 'cnn':
test_data = pad_sentence(test_data, config.filter_sizes[-1])
test_iter, pad_idx = make_iter(config.batch_size,
config.mode,
test_data=test_data)
trainer = Trainer(config, pad_idx, test_iter=test_iter)
trainer.inference()
def main(config):
params_dict = {
'seq2seq': Params('configs/params.json'),
'seq2seq_gru': Params('configs/params_gru.json'),
'seq2seq_attention': Params('configs/params_attention.json')
}
params = params_dict[config.model]
if config.mode == 'train':
train_data, valid_data = load_dataset(config.mode)
train_iter, valid_iter = make_iter(params.batch_size,
config.mode,
train_data=train_data,
valid_data=valid_data)
trainer = Trainer(params,
config.mode,
train_iter=train_iter,
valid_iter=valid_iter)
trainer.train()
else:
test_data = load_dataset(config.mode)
test_iter = make_iter(params.batch_size,
config.mode,
test_data=test_data)
trainer = Trainer(params, config.mode, test_iter=test_iter)
trainer.inference()
def run_pytorch_cnn(train, test, layers, epochs, verbose=True, trials=1, deterministic=True):
if deterministic: make_deterministic()
# Load the dataset
(X_train, y1), (X_val, y2) = train, test
# Add a final dimension indicating the number of channels (only 1 here)
m = X_train.shape[1]
X_train = X_train.reshape((X_train.shape[0], 1, m, m))
X_val = X_val.reshape((X_val.shape[0], 1, m, m))
val_acc, test_acc = 0, 0
for trial in range(trials):
# Reset the weights
for l in layers:
weight_reset(l)
# Make Dataset Iterables
train_iter, val_iter = make_iter(X_train, y1, batch_size=32), make_iter(X_val, y2, batch_size=32)
# Run the model
model, vacc, tacc = \
run_pytorch(train_iter, val_iter, None, layers, epochs, torch.tensor([1,1]), verbose=verbose)
val_acc += vacc if vacc else 0
test_acc += tacc if tacc else 0
if val_acc:
print("\nAvg. validation accuracy:" + str(val_acc / trials))
if test_acc:
print("\nAvg. test accuracy:" + str(test_acc / trials))
predict = model(torch.FloatTensor(X_val)).data.numpy()
return sklearn.metrics.precision_recall_curve(y2, np.amax(predict,axis=1) , pos_label=1)
def run_pytorch_fc(train, test, layers, epochs, verbose=True, trials=1, deterministic=True):
'''
train, test = input data
layers = list of PyTorch layers, e.g. [Linear(in_features=784, out_features=10)]
epochs = number of epochs to run the model for each training trial
trials = number of evaluation trials, resetting weights before each trial
'''
if deterministic:
make_deterministic()
(X_train, y1), (X_val, y2) = train, test
# weights
weights = torch.tensor(sklearn.utils.class_weight.compute_class_weight('balanced',[0,1],y1),dtype=torch.float64)
val_acc, test_acc = 0, 0
for trial in range(trials):
# Reset the weights
for l in layers:
weight_reset(l)
# Make Dataset Iterables
train_iter, val_iter = make_iter(X_train, y1, batch_size=32), make_iter(X_val, y2, batch_size=32)
# Run the model
model, vacc, tacc = \
run_pytorch(train_iter, val_iter, None, layers, epochs, torch.tensor([50,.5]), verbose=verbose)
val_acc += vacc if vacc else 0
test_acc += tacc if tacc else 0
if val_acc:
print("\nAvg. validation accuracy:" + str(val_acc / trials))
if test_acc:
print("\nAvg. test accuracy:" + str(test_acc / trials))
# WARNING: must cast input to FloatTensor not Double Tensor
predict = model(torch.FloatTensor(X_val)).data.numpy()
return sklearn.metrics.precision_recall_curve(y2, np.amax(predict,axis=1) , pos_label=1)
def main(config):
params = Params('config/params.json')
if config.mode == 'train':
train_data, valid_data = load_dataset(config.mode)
train_iter, valid_iter = make_iter(params.batch_size, config.mode, train_data=train_data, valid_data=valid_data)
trainer = Trainer(params, config.mode, train_iter=train_iter, valid_iter=valid_iter)
trainer.train()
else:
test_data = load_dataset(config.mode)
test_iter = make_iter(params.batch_size, config.mode, test_data=test_data)
trainer = Trainer(params, config.mode, test_iter=test_iter)
trainer.inference()