def get_data_label_pair(f, model_config, meta, predict_day, isShift=True):
single_stock = tv_gen._selectData2array_specialDate_v2(
f, corrDate[s][:model_config['corrDate']], model_config['corrDate'],
21, s)
labels = []
data_feature = []
for i in range(model_config['corrDate']):
single_stock_tmp, meta_v = f_extr.create_velocity(
single_stock[i], meta)
single_stock_tmp, meta_ud = f_extr.create_ud_cont_2cls(
single_stock_tmp, meta_v)
features_tmp, label_tmp = dp.get_data_from_normal(
single_stock_tmp, meta_ud, predict_day, model_config['features'],
isShift)
label_tmp = reduce_label(label_tmp)
labels += list(label_tmp)
feature_concat = []
for i in range(model_config['days']):
for k in features_tmp[i]:
feature_concat.append(features_tmp[i][k])
data_feature.append(np.concatenate(feature_concat, axis=1))
data = np.vstack(data_feature)
label = np.array(labels)
return data, label
mConfig = open(
'/home/dashmoment/workspace/etf_prediction/trainer/config/20180526/best_config_xgb_dow_all.pkl',
'rb')
#mConfig = open('/home/ubuntu/shared/workspace/etf_prediction/trainer/config/best_config_xgb_dow_all.pkl', 'rb')
best_config = pickle.load(mConfig)
predict_ud = {}
for s in stock_list:
predict_ud[s] = []
for predict_day in predict_days:
model_config = best_config[s][predict_day]
single_stock = tv_gen._selectData2array(f, [s], model_config['period'])
single_stock, meta_v = f_extr.create_velocity(single_stock, meta)
train_data, train_label = get_data_label_pair(single_stock,
model_config, meta_v)
single_stock_test = tv_gen._selectData2array(f, [s],
['20180401', '20180620'])
single_stock_test, meta_v = f_extr.create_velocity(
single_stock_test, meta)
test_data, test_label = get_data_label_pair(single_stock_test,
model_config, meta_v,
False)
model = model_dict('xgb', model_config).get
model.fit(train_data, train_label)
