def main(num_epochs: int = 100, batch_size: int = 128):
args = {
"news_csv": "data/news_with_splits.csv",
"save_dir": "model_storage/yelp/",
"model_state_file": "model.pth",
"glove_filepath": "data/glove.6B.100d.txt",
"vectorizer_file": "vectorizer.json",
"use_glove": False,
"embedding_size": 100,
"hidden_dim": 100,
"num_channels": 100,
"learning_rate": 0.001,
"num_epochs": num_epochs,
"batch_size": batch_size,
"early_stopping_criteria": 5,
"frequency_cutoff": 25,
"dropout_p": 0.1,
"cuda": False,
}
train_state = make_train_state()
if torch.cuda.is_available():
args["cuda"] = True
args["device"] = torch.device("cuda:0" if args["cuda"] else "cpu")
print(args)
dataset = NewsDataset.load_dataset_and_make_vectorizer(args["news_csv"])
vectorizer = dataset.vectorizer
words = vectorizer.title_vocab._token_to_idx.keys()
embeddings = make_embedding_matrix(glove_filepath=args["glove_filepath"], words=words)
classifier = NewsClassifier(
embedding_size=args["embedding_size"],
num_embeddings=len(vectorizer.title_vocab),
num_channels=args["num_channels"],
hidden_dim=args["hidden_dim"],
num_classes=len(vectorizer.title_vocab),
dropout_p=args["dropout_p"],
pretrained_embeddings=torch.from_numpy(embeddings),
)
classifier = classifier.to(args["device"])
classifier.double()
loss_func = CrossEntropyLoss()
optimizer = Adam(classifier.parameters(), lr=args["learning_rate"])
train(args, train_state, dataset, classifier, optimizer, loss_func, compute_accuracy)
return {
"train_state": train_state,
"args": args,
"dataset": dataset,
"classifier": classifier,
"loss_func": loss_func,
"optimizer": optimizer,
}
def main(batch_size: int = 128, num_epochs: int = 100, hidden_dim: int = 100):
args = {
"hidden_dim": hidden_dim,
"num_channels": 256,
"surname_csv": "data/surnames_with_splits.csv",
"save_dir": "model_storage/yelp/",
"model_state_file": "model.pth",
"vectorizer_file": "vectorizer.json",
"learning_rate": 0.001,
"num_epochs": num_epochs,
"batch_size": batch_size,
"early_stopping_criteria": 5,
"frequency_cutoff": 25,
"cuda": False,
}
train_state = make_train_state()
if torch.cuda.is_available():
args["cuda"] = True
args["device"] = torch.device("cuda:0" if args["cuda"] else "cpu")
print(args)
dataset = SurnameDataset.load_dataset_and_make_vectorizer(
args["surname_csv"], SurnameVectorizer.from_dataframe)
vectorizer = dataset.vectorizer
classifier = SurnameCnnClassifier(
initial_num_channels=len(vectorizer.surname_vocab),
num_classes=len(vectorizer.nationality_vocab),
num_channels=args["num_channels"],
)
classifier = classifier.to(args["device"])
loss_func = CrossEntropyLoss(dataset.class_weights)
optimizer = Adam(classifier.parameters(), lr=args["learning_rate"])
train(args, train_state, dataset, classifier, optimizer, loss_func,
compute_accuracy)
return {
"train_state": train_state,
"args": args,
"dataset": dataset,
"classifier": classifier,
"loss_func": loss_func,
"optimizer": optimizer,
}
def main(num_epochs: int = 100, batch_size: int = 128):
args = {
"cbow_csv": "data/frankenstein_with_splits.csv",
"save_dir": "model_storage/yelp/",
"model_state_file": "model.pth",
"vectorizer_file": "vectorizer.json",
"embedding_size": 300,
"learning_rate": 0.001,
"num_epochs": num_epochs,
"batch_size": batch_size,
"early_stopping_criteria": 5,
"frequency_cutoff": 25,
"cuda": False,
}
train_state = make_train_state()
if torch.cuda.is_available():
args["cuda"] = True
args["device"] = torch.device("cuda:0" if args["cuda"] else "cpu")
print(args)
dataset = CbowDataset.load_dataset_and_make_vectorizer(args["cbow_csv"])
vectorizer = dataset.vectorizer
classifier = CbowClassifier(
vocabulary_size=len(vectorizer.cbow_vocab),
embedding_size=args["embedding_size"],
)
classifier = classifier.to(args["device"])
loss_func = CrossEntropyLoss()
optimizer = Adam(classifier.parameters(), lr=args["learning_rate"])
train(args, train_state, dataset, classifier, optimizer, loss_func,
compute_accuracy)
return {
"train_state": train_state,
"args": args,
"dataset": dataset,
"classifier": classifier,
"loss_func": loss_func,
"optimizer": optimizer,
}
def main(batch_size: int = 128, num_epochs: int = 100):
args = {
"review_csv": "data/yelp_reviews_lite.json",
"save_dir": "model_storage/yelp/",
"model_state_file": "model.pth",
"vectorizer_file": "vectorizer.json",
"learning_rate": 0.001,
"num_epochs": num_epochs,
"batch_size": batch_size,
"early_stopping_criteria": 5,
"frequency_cutoff": 25,
"cuda": False,
}
train_state = make_train_state()
if torch.cuda.is_available():
args["cuda"] = True
args["device"] = torch.device("cuda:0" if args["cuda"] else "cpu")
print(args)
dataset = ReviewDataset.load_dataset_and_make_vectorizer(
args["review_csv"])
vectorizer = dataset.vectorizer
classifier = ReviewClassifier(num_features=len(vectorizer.review_vocab))
classifier = classifier.to(args["device"])
loss_func = nn.BCEWithLogitsLoss()
optimizer = optim.Adam(classifier.parameters(), lr=args["learning_rate"])
train(args, train_state, dataset, classifier, optimizer, loss_func,
compute_accuracy)
return {
"train_state": train_state,
"args": args,
"dataset": dataset,
"classifier": classifier,
"loss_func": loss_func,
"optimizer": optimizer,
}