df_train, df_test = data.load_data(cache=True)
df = data.create_fulldf(df_train, df_test)
df = df.fillna(NULL_VALUE)
df = data.clean_data(df)
df = data.encode_labels(df)
#df = features.add_features(df)
logerror = df['logerror'].values
targets = logerror
df = data.select_features(df)
print df.columns
if cv_flag:
df_full_train, targets, df_test = data.split_data(df, logerror)
df_train, df_test, train_targets, test_targets = data.split_cv(
df_full_train, targets, cv_split_ratio)
dtest = xgb.DMatrix(df_test.values, test_targets)
dtrain = xgb.DMatrix(df_train.values, train_targets)
params = model_params.get_xtune11k()
cv_preds = np.repeat(0, len(df_test))
num_boost_rounds = 110
for i in range(n_bags):
watchlist = [(dtrain, 'train'), (dtest, 'eval')]
model = xgb.train(params,
dtrain,
num_boost_round=num_boost_rounds,
evals=watchlist,
early_stopping_rounds=50)
cv_preds = model.predict(dtest) + cv_preds
df = data.create_fulldf(df_train, df_test)
df = df.fillna(NULL_VALUE)
df = data.clean_data(df)
df = data.encode_labels(df)
#df = features.add_features(df)
logerror = df['logerror'].values
targets = logerror
df = data.select_features(df)
df = df.drop(['assessmentyear'], axis=1)
print df.columns
if cv_flag:
df_full_train, targets, df_test = data.split_data(df, logerror)
df_train, df_val, train_targets, val_targets = data.split_cv(df_full_train, targets, cv_split_ratio)
cv_preds = np.repeat(0., len(df_val))
for i in range(n_bags):
x_train, x_val = tools.normalise_data(df_train.values, df_val.values)
model = model_params.get_keras(x_train.shape[1])
history = model.fit(
x_train, train_targets,
nb_epoch=epochs, batch_size=batch_size,
validation_data=(x_val, val_targets), verbose=2)
model.history = history
cv_preds += model.predict(x_val).squeeze()
cv_preds /= float(n_bags)
mae = tools.get_mae_loss(val_targets, cv_preds)
mse = mean_squared_error(val_targets, cv_preds)
if __name__ == '__main__':
df_train, df_test = data.load_data(cache=True)
df = data.create_fulldf(df_train, df_test)
df = df.fillna(NULL_VALUE)
df = data.clean_data(df)
df = data.encode_labels(df)
#df = features.add_features(df)
targets = df['logerror'].values
df = data.select_features(df)
print df.columns
df_full_train, targets, df_test = data.split_data(df, targets)
df_train, df_test, train_targets, test_targets = data.split_cv(
df_full_train, targets, 0.8)
dtrain = xgb.DMatrix(df_train.values, train_targets)
dtest = xgb.DMatrix(df_test.values, test_targets)
y_mean = np.mean(train_targets)
watchlist = [(dtrain, 'train'), (dtest, 'eval')]
num_boost_rounds = 80000
while True:
params = sample_params(random=False)
model = xgb.train(params,
dtrain,
num_boost_round=num_boost_rounds,
evals=watchlist,
early_stopping_rounds=15)