def train():
space = "words"
train_df = DataSet.load_train()
xtr_df, xval_df = train_test_split(train_df, test_size=0.25)
test_df = DataSet.load_test()
### Generate data generator
train_dg = DataGenerator(data_df=xtr_df,space=space,bucket_num=5,batch_size=5000,is_prefix_pad=True,is_shuffle=True,is_test=False)
val_dg = DataGenerator(data_df=xval_df,space=space,bucket_num=5,batch_size=5000,is_prefix_pad=True,is_shuffle=False,is_test=False)
test_dg = DataGenerator(data_df=test_df,space=space,bucket_num=5,batch_size=5000,is_prefix_pad=True,is_shuffle=False,is_test=True)
### Must do prepare before using
train_dg.prepare()
val_dg.prepare()
test_dg.prepare()
### load word embedding, can use train_df, val_dg or test_dg
item_embed = train_dg.get_item_embed_tensor(space)
### Initialize network
siamese_cnn = Siamese_CNN(item_embed,is_freeze=True)
### Initialize model using network
siamese_model = Model(siamese_cnn)
criteria = nn.BCEWithLogitsLoss()
optimizer_ft = optim.Adam(ifilter(lambda p: p.requires_grad, siamese_cnn.parameters()), lr=8e-4)
exp_lr_scheduler = lr_scheduler.ExponentialLR(optimizer_ft, gamma=0.9)
### Train
siamese_model.train(train_dg,val_dg,criteria,optimizer_ft,exp_lr_scheduler,25)
preds = siamese_model.predict(test_dg).numpy()
preds = pd.DataFrame({"y_pre":preds})
preds.to_csv("submission.csv",index=False)
def cv_main():
kf = KFold(n_splits=folder, shuffle=True, random_state=19920618)
all_train_df = DataSet.load_train()
test_df = DataSet.load_test()
test_dg = DataGenerator(data_df=test_df,
space=space,
bucket_num=5,
batch_size=256,
is_prefix_pad=False,
is_shuffle=False,
is_test=True)
print("prepare test data generator")
test_dg.prepare()
item_embed = test_dg.get_item_embed_tensor(space)
train_eval = np.zeros(len(all_train_df))
test_eval = np.zeros((len(test_df), folder))
for i, (train_index, val_index) in enumerate(kf.split(all_train_df)):
print()
train_name = version + "_cv_%s" % (i)
xtr_df = all_train_df.iloc[train_index]
xval_df = all_train_df.iloc[val_index]
train_dg = DataGenerator(data_df=xtr_df,
space=space,
bucket_num=5,
batch_size=batch_size,
is_prefix_pad=False,
is_shuffle=True,
is_test=False)
val_dg = DataGenerator(data_df=xval_df,
space=space,
bucket_num=5,
batch_size=256,
is_prefix_pad=False,
is_shuffle=False,
is_test=False)
print("prepare train data generator, cv_%s" % i)
train_dg.prepare()
print("prepare val data generator, cv_%s" % i)
val_dg.prepare()
siamese_lstm = Siamese_LSTM(pre_trained_embedding=item_embed,
is_freeze=is_freeze,
hidden_size=hidden_size,
number_layers=num_layers,
lstm_dropout_p=lstm_drop_p,
bidirectional=bidirectional,
linear_hid_size=linear_hidden_size,
linear_hid_drop_p=linear_hid_drop_p,
input_drop_p=lstm_input_drop_p)
siamese_lstm.init_weights(
) ##TODO Whether to initialize customised weights as Keras
siamese_model = Model(train_name, siamese_lstm)
criteria = nn.BCEWithLogitsLoss()
optimizer_ft = optim.Adam(ifilter(lambda p: p.requires_grad,
siamese_lstm.parameters()),
lr=LR)
exp_lr_scheduler = lr_scheduler.ExponentialLR(optimizer_ft,
gamma=Gamma)
### Train
siamese_model.train(train_dg=train_dg,
valid_dg=val_dg,
criterion=criteria,
optimizer=optimizer_ft,
scheduler=exp_lr_scheduler,
num_epochs=num_epochs,
early_stop_rounds=early_stop)
siamese_model.save_plot_()
val_pred = siamese_model.predict(val_dg).numpy()
train_eval[val_index] = val_pred
test_preds = siamese_model.predict(test_dg).numpy()
test_eval[:, i] = test_preds
train_pred_df = pd.DataFrame({version + "_train_pred_cv": train_eval})
train_pred_df.to_csv(version + "_train_pred_cv.csv", index=False)
test_pred_df = pd.DataFrame(
test_eval,
columns=[version + "_test_pred_cv_%s" % (i) for i in xrange(folder)])
test_pred_df["y_pre"] = test_pred_df.mean(axis=1)
test_pred_df.to_csv(version + "_test_pred_cv.csv", index=False)
test_pred_df[["y_pre"]].to_csv(version + "_submission_cv.csv", index=False)
def train_main():
train_name = version + "_sm"
##--------------parameters-------------------##
train_df = DataSet.load_train()
xtr_df, xval_df = train_test_split(train_df, test_size=0.20)
test_df = DataSet.load_test()
### Generate data generator
train_dg = DataGenerator(data_df=xtr_df,
space=space,
bucket_num=5,
batch_size=batch_size,
is_prefix_pad=False,
is_shuffle=True,
is_test=False)
val_dg = DataGenerator(data_df=xval_df,
space=space,
bucket_num=5,
batch_size=512,
is_prefix_pad=False,
is_shuffle=False,
is_test=False)
test_dg = DataGenerator(data_df=test_df,
space=space,
bucket_num=5,
batch_size=512,
is_prefix_pad=False,
is_shuffle=False,
is_test=True)
### Must do prepare before using
train_dg.prepare()
val_dg.prepare()
test_dg.prepare()
### load word embedding, can use train_df, val_dg or test_dg
item_embed = train_dg.get_item_embed_tensor(space)
### Initialize network
siamese_lstm = Siamese_LSTM(pre_trained_embedding=item_embed,
is_freeze=is_freeze,
hidden_size=hidden_size,
number_layers=num_layers,
lstm_dropout_p=lstm_drop_p,
bidirectional=bidirectional,
linear_hid_size=linear_hidden_size,
linear_hid_drop_p=linear_hid_drop_p,
input_drop_p=lstm_input_drop_p)
siamese_lstm.init_weights(
) ##TODO Whether initialize customised weights as Keras
### Initialize model using network
siamese_model = Model(train_name, siamese_lstm)
criteria = nn.BCEWithLogitsLoss()
optimizer_ft = optim.Adam(ifilter(lambda p: p.requires_grad,
siamese_lstm.parameters()),
lr=LR)
exp_lr_scheduler = lr_scheduler.ExponentialLR(optimizer_ft, gamma=Gamma)
### Train
siamese_model.train(train_dg=train_dg,
valid_dg=val_dg,
criterion=criteria,
optimizer=optimizer_ft,
scheduler=exp_lr_scheduler,
num_epochs=num_epochs,
early_stop_rounds=early_stop)
siamese_model.save_plot_()
preds = siamese_model.predict(test_dg).numpy()
preds = pd.DataFrame({"y_pre": preds})
preds.to_csv(version + "_submission_sm.csv", index=False)
def train():
space = "words"
is_freeze = True
hidden_size = 100
num_layers = 2
lstm_dropput_p = 0.6 ##TODO 0.4->0.5->0.6
lstm_input_dropout = 0.6
bidirectional = True
linear_hidden_size = 200
linear_hid_drop_p = 0.3
train_name = "v0.2"
train_df = DataSet.load_train()
xtr_df, xval_df = train_test_split(train_df, test_size=0.25)
test_df = DataSet.load_test()
### Generate data generator
train_dg = DataGenerator(data_df=xtr_df,
space=space,
bucket_num=5,
batch_size=512,
is_prefix_pad=True,
is_shuffle=True,
is_test=False)
val_dg = DataGenerator(data_df=xval_df,
space=space,
bucket_num=5,
batch_size=512,
is_prefix_pad=True,
is_shuffle=False,
is_test=False)
test_dg = DataGenerator(data_df=test_df,
space=space,
bucket_num=5,
batch_size=512,
is_prefix_pad=True,
is_shuffle=False,
is_test=True)
### Must do prepare before using
train_dg.prepare()
val_dg.prepare()
test_dg.prepare()
### load word embedding, can use train_df, val_dg or test_dg
item_embed = train_dg.get_item_embed_tensor(space)
### Initialize network
siamese_lstm = Siamese_LSTM(pre_trained_embedding=item_embed,
is_freeze=is_freeze,
hidden_size=hidden_size,
number_layers=num_layers,
lstm_dropout_p=lstm_dropput_p,
bidirectional=bidirectional,
linear_hid_size=linear_hidden_size,
linear_hid_drop_p=linear_hid_drop_p,
input_drop_p=lstm_input_dropout)
### Initialize model using network
siamese_model = Model(siamese_lstm)
criteria = nn.BCEWithLogitsLoss()
optimizer_ft = optim.Adam(ifilter(lambda p: p.requires_grad,
siamese_lstm.parameters()),
lr=0.001) ##TODO 0.001
exp_lr_scheduler = lr_scheduler.ExponentialLR(optimizer_ft,
gamma=0.99) ##TODO 0.99
### Train
siamese_model.train(train_name, train_dg, val_dg, criteria, optimizer_ft,
exp_lr_scheduler, 150) ##TODO 150
preds = siamese_model.predict(test_dg).numpy()
preds = pd.DataFrame({"y_pre": preds})
preds.to_csv("submission.csv", index=False)
