top_param=3)
top_rank_5 = model.top_rank_accuracy(prediction,
validation_dataset,
top_param=5)
loss = model.loss_fn(prediction, outputs)
print(f"Validation Loss : {loss.data[0]}")
return loss.data[0], top_rank_1, top_rank_3, top_rank_5
if __name__ == "__main__":
use_cuda = torch.cuda.is_available()
dataset = SimpleDataset(
training_file="IR_train_hard.json",
preprocessing=False,
preprocessed_data_filename="hard_training_unprocessed")
validation_dataset = SimpleDataset(
training_file="IR_val_hard.json",
preprocessing=False,
preprocessed_data_filename="hard_val_unprocessed")
model, top_rank_1_arr, \
top_rank_3_arr, top_rank_5_arr = train_cbow_reg_network(
dataset,
validation_dataset,
num_epochs=30,
batch_size=256,
embedding_space=300,
hidden_layer_dim=256,
save_model(f"{modelname}_Easy", hidden_layer_dim, embedding_space,
learning_rate, loss_fn_name, model)
return top_rank_1_arr[-1], top_rank_3_arr[-1], top_rank_5_arr[-1]
else:
loss, top_rank_1, \
top_rank_3, top_rank_5 = validate_model(dataset.vocab_size, dataset.w2i, validation_dataset, model=model)
return top_rank_1, top_rank_3, top_rank_5
if __name__ == '__main__':
use_cuda = torch.cuda.is_available()
easy_dataset = SimpleDataset(
training_file="IR_train_easy.json",
preprocessing=True,
preprocessed_data_filename="easy_training_processed_with_questions")
test_dataset = SimpleDataset(
training_file="IR_test_easy.json",
preprocessing=True,
preprocessed_data_filename="easy_val_processed_with_questions")
loss_fn = torch.nn.MSELoss(size_average=True)
loss_fn = torch.nn.SmoothL1Loss(size_average=True)
Embedding_Spaces = [100, 150, 200, 250, 300, 350]
Hidden_Dims = [56, 256, 512, 1024]
Learning_Rates = [0.001, 0.0001, 0.00001]
best_top_1 = 0
best_top_params = []
top_rank_1 = model.top_rank_accuracy(prediction, outputs, top_param=1)
top_rank_3 = model.top_rank_accuracy(prediction, outputs, top_param=3)
top_rank_5 = model.top_rank_accuracy(prediction, outputs, top_param=5)
loss = model.loss_fn(prediction, outputs)
print(f"Validation Loss : {loss.data[0]}")
return loss.data[0], top_rank_1, top_rank_3, top_rank_5
if __name__ == "__main__":
use_cuda = torch.cuda.is_available()
dataset = SimpleDataset(
training_file="IR_train_easy.json",
preprocessing=True,
preprocessed_data_filename="easy_training_processed")
validation_dataset = SimpleDataset(
training_file="IR_val_easy.json",
preprocessing=True,
preprocessed_data_filename="easy_val_processed")
model, top_rank_1_arr, \
top_rank_3_arr, top_rank_5_arr = train_cbow_network(
dataset,
validation_dataset,
num_epochs=30,
batch_size=64,
embedding_space=150,
hidden_layer_dim=256,
model.load_state_dict(torch.load(f"models/{model_hard_file}"))
loss, top_rank_1, \
top_rank_3, top_rank_5 = validation_func(hard_dataset.vocab_size, hard_dataset.w2i, hard_testset, model=model)
print("HARD RESULTS")
print(top_rank_1)
print(top_rank_3)
print(top_rank_5)
if __name__ == '__main__':
easy_dataset = SimpleDataset(
training_file="IR_train_easy.json",
preprocessing=True,
preprocessed_data_filename="easy_training_processed_with_questions"
)
easy_dataset_cbow = SimpleDataset(
training_file="IR_train_easy.json",
preprocessing=True,
preprocessed_data_filename="easy_training_unprocessed" # Saved to wrong file, didn't want to wait an hour to reprocess
)
easy_test_dataset = SimpleDataset(
training_file="IR_test_easy.json",
preprocessing=True,
preprocessed_data_filename="easy_test_processed_with_questions"
)