def train(x_idx, x_value, label, features):
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
X_idx_tensor = torch.LongTensor(x_idx).to(device)
X_value_tensor = torch.Tensor(x_value).to(device)
y_tensor = torch.Tensor(label).to(device)
y_tensor = y_tensor.reshape(-1, 1)
X = TensorDataset(X_idx_tensor, X_value_tensor, y_tensor)
model = xDeepFM(features.feature_size(),
features.field_size(),
dropout_deep=[0, 0, 0],
deep_layer_sizes=[400, 400],
cin_layer_sizes=[100, 100, 50],
embedding_size=6).to(device)
optimizer = torch.optim.Adam(model.parameters())
model_path = os.path.join(pwd_path, 'xdeepfm_model.pt')
model, loss_history = train_model(model=model,
model_path=model_path,
dataset=X,
loss_func=nn.BCELoss(),
optimizer=optimizer,
device=device,
val_size=0.2,
batch_size=32,
epochs=40,
patience=10)
print(loss_history)
def train(x_idx, x_value, label, features, out_type='binary'):
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
X_idx_tensor = torch.LongTensor(x_idx).to(device)
X_value_tensor = torch.Tensor(x_value).to(device)
y_tensor = torch.Tensor(label).to(device)
y_tensor = y_tensor.reshape(-1, 1)
X = TensorDataset(X_idx_tensor, X_value_tensor, y_tensor)
model = DeepFM(feature_size=features.feature_size(),
field_size=features.field_size(),
out_type=out_type).to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
model_path = os.path.join(pwd_path, 'deepfm_model.pt')
model, loss_history = train_model(model=model,
model_path=model_path,
dataset=X,
loss_func=nn.BCELoss(),
optimizer=optimizer,
device=device,
val_size=0.2,
batch_size=32,
epochs=10,
shuffle=True)
print(loss_history)
def train(x_idx, x_value, label):
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
X_idx_tensor = torch.LongTensor(x_idx).to(device)
X_value_tensor = torch.Tensor(x_value).to(device)
y_tensor = torch.Tensor(label).to(device)
y_tensor = y_tensor.reshape(-1, 1)
X = TensorDataset(X_idx_tensor, y_tensor)
model = GBDTLR().to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
model_path = os.path.join(pwd_path, 'gbdtlr_model.pt')
model.train_gbdt(x_idx.values.tolist(), label.values.tolist())
model, loss_history = train_model(model=model,
model_path=model_path,
dataset=X,
loss_func=nn.BCELoss(),
optimizer=optimizer,
device=device,
val_size=0.2,
batch_size=32,
epochs=10)
print(loss_history)
def train(x_idx,
x_value,
label,
features,
categorical_index,
continuous_value,
out_type='binary'):
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
X_idx_tensor = torch.LongTensor(x_idx).to(device)
X_value_tensor = torch.Tensor(x_value).to(device)
y_tensor = torch.Tensor(label).to(device)
y_tensor = y_tensor.reshape(-1, 1)
continuous_value = torch.FloatTensor(continuous_value).to(device)
categorical_index = torch.FloatTensor(categorical_index).to(device)
X = TensorDataset(continuous_value, categorical_index, y_tensor)
model = WideAndDeep(feature_size=features.feature_size(),
categorical_field_size=26,
continuous_field_size=13,
out_type=out_type).to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
model_path = os.path.join(pwd_path, 'wad_model.pt')
model, loss_history = train_model(model=model,
model_path=model_path,
dataset=X,
loss_func=nn.BCELoss(),
optimizer=optimizer,
device=device,
val_size=0.2,
batch_size=32,
epochs=10,
shuffle=True)
print(loss_history)
