print(train.columns)
X_train = train[cols_to_use].values
y_train = train['click'].values
# X_train = X_train[:100]
# y_train = y_train[:100]
testID = test['ID']
test = test[cols_to_use].values
print(X_train.shape, y_train.shape)
print("Saving data.....")
ls.save_obj(X_train, "X_train")
ls.save_obj(y_train, "y_train")
ls.save_obj(testID, "testID")
ls.save_obj(test, "test")
exit()
print("Training data.....")
X_train = ls.load_obj("X_train")
y_train = ls.load_obj("y_train")
testID = ls.load_obj("testID")
test = ls.load_obj("test")
click_weight = class_weight.compute_class_weight('balanced',
np.unique(y_train), y_train)
def auc(y_true, y_pred):
return metrics.roc_auc_score(y_true, y_pred)
def get_class_weights(y, smooth_factor=0):
"""
values around 0.1 (smooth factor) are a good default for very imbalanced classes.
Returns the weights for each class based on the frequencies of the samples
print(train.columns)
X_train = train[cols_to_use].values
y_train = train['click'].values
# X_train = X_train[:100]
# y_train = y_train[:100]
testID = test['ID']
test = test[cols_to_use].values
print(X_train.shape, y_train.shape)
print("Saving data.....")
ls.save_obj(X_train, "X_train_lstm")
ls.save_obj(y_train, "y_train_lstm")
ls.save_obj(testID, "testID_lstm")
ls.save_obj(test, "test_lstm")
exit()
print("Training data.....")
X_train = ls.load_obj("X_train_lstm")
y_train = ls.load_obj("y_train_lstm")
testID = ls.load_obj("testID_lstm")
test = ls.load_obj("test_lstm")
testUnique = np.unique(test)
trainUnique = np.unique(X_train)
dataUnique = np.unique(np.concatenate([trainUnique, testUnique]))
print("X_train", X_train.shape)
print("testUnique", testUnique.shape)
print("trainUnique", trainUnique.shape)
print("dataUnique", dataUnique.shape)
max_features = dataUnique.shape[0]
click_weight = class_weight.compute_class_weight('balanced',
np.unique(y_train), y_train)