def regressor_dev_test(self, dev_set, dev_value_set, dev_date_set, dev_stock_id_set, save_clsfy_path="mlp_trade_regressor", is_cv=False, include_top_list = None):
# test mode
if not include_top_list:
include_top_list = [1]
mlp_regressor = self.mlp_regressor
pred_value_list = np.array(mlp_regressor.predict(dev_set))
actual_value_list = np.array(dev_value_set)
mrse = calculate_rmse(actual_value_list, pred_value_list)
date_list = dev_date_set
stock_id_list = dev_stock_id_set
avg_price_change_tuple, var_tuple, std_tuple = get_avg_price_change(pred_value_list, actual_value_list,
date_list, stock_id_list,
include_top_list=
include_top_list)
# count how many predicted value has the same polarity as actual value
polar_list = [1 for x, y in zip(pred_value_list, actual_value_list) if x * y >= 0]
polar_count = len(polar_list)
polar_percent = polar_count / len(pred_value_list)
#
# <uncomment for debugging>
if not is_cv:
print("----------------------------------------------------------------------------------------")
print("actual_value_list, ", actual_value_list)
print("pred_value_list, ", pred_value_list)
print("polarity: {}".format(polar_percent))
print("mrse: {}".format(mrse))
print("avg_price_change: {}".format(avg_price_change_tuple))
print("----------------------------------------------------------------------------------------")
else:
pass
# print("Testing complete! Testing Set size: {}".format(len(self.r_dev_value_set)))
# <uncomment for debugging>
# ------------------------------------------------------------------------------------------------------------------
return mrse, avg_price_change_tuple[0], polar_percent
def regressor_dev(self, save_clsfy_path="mlp_trade_regressor", is_cv=False, include_top_list = None):
# test mode
if not include_top_list:
include_top_list = [1]
mlp_regressor = pickle.load(open(save_clsfy_path, "rb"))
pred_value_list = np.array(mlp_regressor.predict(self.dev_set))
actual_value_list = np.array(self.dev_value_set)
date_list = self.dev_date_set
stock_id_list = self.dev_stock_id_set
avg_price_change_tuple, var_tuple, std_tuple = get_avg_price_change(pred_value_list, actual_value_list,
date_list, stock_id_list,
include_top_list=
include_top_list)
# compute accuracy in terms of positive and negative
# (3.) get the pred label for each week
pred_label_dict_by_week = collections.defaultdict(lambda :[])
golden_label_dict_by_week = collections.defaultdict(lambda :[])
pred_value_dict_by_week = collections.defaultdict(lambda :[])
golden_value_dict_by_week = collections.defaultdict(lambda :[])
pred_label_list = ['pos' if x >= 0 else 'neg' for x in pred_value_list ]
actual_label_list = ['pos' if x >= 0 else 'neg' for x in actual_value_list ]
for i, pred_label in enumerate(pred_label_list):
date = self.dev_date_set[i]
# classification
pred_label_dict_by_week[date].append(pred_label)
golden_label = actual_label_list[i]
golden_label_dict_by_week[date].append(golden_label)
#
# regression
predict_value = pred_value_list[i]
golden_value = actual_value_list[i]
pred_value_dict_by_week[date].append(predict_value)
golden_value_dict_by_week[date].append(golden_value)
week_average_f1_list = []
week_average_accuracy_list = []
week_average_rmse = []
dev_label_dict = collections.defaultdict(lambda: 0)
pred_label_dict = collections.defaultdict(lambda: 0)
label_f1_list_all = []
# (4.) compute the f1, accuracy for each week in 1 validation set
for date, pred_label_list_for_1_week in pred_label_dict_by_week.items():
pred_label_list = pred_label_list_for_1_week
golden_label_list = golden_label_dict_by_week[date]
# (3.) compute the average f-measure
_,average_f1 = compute_average_f1(pred_label_list, golden_label_list)
week_average_f1_list.append(average_f1)
#average_f1 = f1_list[0] # using F-measure
#
# (4.) compute accuracy
correct = 0
for i, pred_label in enumerate(pred_label_list):
if pred_label == golden_label_list[i]:
correct += 1
accuracy = correct / len(golden_label_list)
week_average_accuracy_list.append(accuracy)
#
# (5.) count the occurrence for each label
for dev_label in golden_label_list:
dev_label_dict[dev_label] += 1
for pred_label in pred_label_list:
pred_label_dict[pred_label] += 1
#
# # (6.) save rmse
pred_value_list1 = pred_value_dict_by_week[date]
actual_value_list1 = golden_value_dict_by_week[date]
rmse = calculate_rmse(actual_value_list1, pred_value_list1)
week_average_rmse.append(rmse)
week_average_f1 = np.average(week_average_f1_list)
week_average_accuracy = np.average(week_average_accuracy_list)
week_average_rmse = np.average(week_average_rmse)
# <uncomment for debugging>
if not is_cv:
print("----------------------------------------------------------------------------------------")
print("actual_value_list, ", actual_value_list)
print("pred_value_list, ", pred_value_list)
print("week_average_accuracy: {}".format(week_average_accuracy))
print("week_average_f1: {}".format(week_average_f1))
print("week_average_rmse: {}".format(week_average_rmse))
print("week_average_price_change: {}".format(avg_price_change_tuple))
print("----------------------------------------------------------------------------------------")
else:
pass
# print("Testing complete! Testing Set size: {}".format(len(self.r_dev_value_set)))
# <uncomment for debugging>
# ------------------------------------------------------------------------------------------------------------------
return week_average_rmse, avg_price_change_tuple, week_average_accuracy, week_average_f1
def baseline_reg_dev(self, target_folder):
file1 = os.listdir(target_folder)[0]
file1_path = os.path.join(target_folder, file1)
with open(file1_path, 'r') as f:
feature_list = f.readlines()[0].strip().split(',')[::2]
key_index = feature_list.index('percent_change_price')
print ("key_index: ", key_index)
pred_value_list = []
for dev_sample in self.dev_set:
percent_change_price = float(dev_sample[key_index])*0.01
pred_value_list.append(percent_change_price)
actual_value_list = np.array(self.dev_value_set)
date_list = self.dev_date_set
stock_id_list = self.dev_stock_id_set
include_top_list = [1]
avg_price_change_tuple, var_tuple, std_tuple = get_avg_price_change(pred_value_list, actual_value_list,
date_list, stock_id_list,
include_top_list=
include_top_list)
date_actual_avg_priceChange_list = get_chosen_stock_return(pred_value_list, actual_value_list, date_list,
stock_id_list, include_top_list=None)
# compute accuracy in terms of positive and negative
# (3.) get the pred label for each week
pred_label_dict_by_week = collections.defaultdict(lambda :[])
golden_label_dict_by_week = collections.defaultdict(lambda :[])
pred_value_dict_by_week = collections.defaultdict(lambda :[])
golden_value_dict_by_week = collections.defaultdict(lambda :[])
pred_label_list = ['pos' if x >= 0 else 'neg' for x in pred_value_list ]
actual_label_list = ['pos' if x >= 0 else 'neg' for x in actual_value_list ]
for i, pred_label in enumerate(pred_label_list):
date = self.dev_date_set[i]
# classification
pred_label_dict_by_week[date].append(pred_label)
golden_label = actual_label_list[i]
golden_label_dict_by_week[date].append(golden_label)
#
# regression
predict_value = pred_value_list[i]
golden_value = actual_value_list[i]
pred_value_dict_by_week[date].append(predict_value)
golden_value_dict_by_week[date].append(golden_value)
week_average_f1_list = []
week_average_accuracy_list = []
week_average_rmse = []
dev_label_dict = collections.defaultdict(lambda: 0)
pred_label_dict = collections.defaultdict(lambda: 0)
label_f1_list_all = []
# (4.) compute the f1, accuracy for each week in 1 validation set
for date, pred_label_list_for_1_week in pred_label_dict_by_week.items():
pred_label_list = pred_label_list_for_1_week
golden_label_list = golden_label_dict_by_week[date]
# (3.) compute the average f-measure
_,average_f1 = compute_average_f1(pred_label_list, golden_label_list)
week_average_f1_list.append(average_f1)
#average_f1 = f1_list[0] # using F-measure
#
# (4.) compute accuracy
correct = 0
for i, pred_label in enumerate(pred_label_list):
if pred_label == golden_label_list[i]:
correct += 1
accuracy = correct / len(golden_label_list)
week_average_accuracy_list.append(accuracy)
#
# (5.) count the occurrence for each label
for dev_label in golden_label_list:
dev_label_dict[dev_label] += 1
for pred_label in pred_label_list:
pred_label_dict[pred_label] += 1
#
# # (6.) save rmse
pred_value_list1 = pred_value_dict_by_week[date]
actual_value_list1 = golden_value_dict_by_week[date]
rmse = calculate_rmse(actual_value_list1, pred_value_list1)
week_average_rmse.append(rmse)
week_average_f1 = np.average(week_average_f1_list)
week_average_accuracy = np.average(week_average_accuracy_list)
week_average_rmse = np.average(week_average_rmse)
# # <uncomment for debugging>
# print("----------------------------------------------------------------------------------------")
# print("actual_value_list, ", actual_value_list)
# print("pred_value_list, ", pred_value_list)
# print("week_average_accuracy: {}".format(week_average_accuracy))
# print("week_average_f1: {}".format(week_average_f1))
# print("week_average_rmse: {}".format(week_average_rmse))
# print("week_average_price_change: {}".format(avg_price_change_tuple))
# print("----------------------------------------------------------------------------------------")
# ------------------------------------------------------------------------------------------------------------------
return week_average_rmse, avg_price_change_tuple, week_average_accuracy, week_average_f1, \
date_actual_avg_priceChange_list
rnn1.train_rnn()
pred_value_list, actual_value_list, date_list, stock_id_list = rnn1.reg_dev_for_moving_window_test(
)
#
#print ("date_list: ", date_list)
pred_label_list_temp = [
'pos' if x >= 0 else 'neg' for x in pred_value_list
]
actual_label_list_temp = [
'pos' if x >= 0 else 'neg' for x in actual_value_list
]
pred_label_list.extend(pred_label_list_temp)
actual_label_list.extend(actual_label_list_temp)
data_list_for_classification.extend(date_list)
#
rmse = calculate_rmse(actual_value_list, pred_value_list)
avg_price_change_tuple, var_tuple, std_tuple = get_avg_price_change(
pred_value_list, actual_value_list, date_list, stock_id_list)
chosen_stock_return_list_temp = get_chosen_stock_return(
pred_value_list, actual_value_list, date_list, stock_id_list)
avg_price_change_1 = avg_price_change_tuple[
0] # Strategy: choose the top 1 stock each week
var_1 = var_tuple[0]
std_1 = std_tuple[0]
avg_price_change_list.append(avg_price_change_1)
print("chosen_stock_return_list_temp: ", chosen_stock_return_list_temp)
var_list.append(var_1)
std_list.append(std_1)
chosen_stock_return_list.extend(chosen_stock_return_list_temp)
rmse_list.append(rmse)
def stock_prediction_baseline_reg(train_data_folder_name,
test_data_folder_name,
is_random=False,
is_highest_profit=False,
random_seed=None,
data_set='a_share'):
print("Baseline for regression!")
# (1.) build classifer
mlp_classifier1 = MlpTradeRegressor()
#hidden_layer_sizes = (26,6)
#learning_rate_init = 0.0001
#print ("hidden_layer_sizes: ", hidden_layer_sizes)
##mlp_classifier1.set_regressor(hidden_layer_sizes, learning_rate_init = learning_rate_init)
#clsfy_name = 'dow_jones_mlp_trade_classifier_window_shift'
#clf_path = os.path.join(parent_folder, 'trained_classifiers', clsfy_name)
# (2.) data folder
train_data_folder = os.path.join(data_set, train_data_folder_name)
train_data_folder = os.path.join(parent_folder, 'data', train_data_folder)
test_data_folder = os.path.join(data_set, test_data_folder_name)
test_data_folder = os.path.join(parent_folder, 'data', test_data_folder)
#
# read train data set
train_data_set = set()
train_stock_set = set()
train_baseline_dict = collections.defaultdict(
lambda: collections.defaultdict(lambda: collections.defaultdict(lambda:
0)))
file_name_list = os.listdir(train_data_folder)
for file_name in file_name_list:
data_str = re.findall(r'([0-9\-]+)_', file_name)[0]
stock_name = re.findall(r'[0-9\-]+_([A-Za-z0-9]+)_', file_name)[0]
price_change = float(re.findall(r'#([0-9\.\-]+)#', file_name)[0])
train_baseline_dict[data_str][stock_name]['actual'] = price_change
train_data_set.add(data_str)
train_stock_set.add(stock_name)
train_data_list = sorted(list(train_data_set))
print("train_data_list: ", train_data_list)
print("train_stock_set: ", train_stock_set)
#
# read test data set
test_date_set = set()
test_stock_set = set()
NUMBER_OF_PREVIOUS_WEEK = 2
baseline_dict = collections.defaultdict(lambda: collections.defaultdict(
lambda: collections.defaultdict(lambda: 0)))
file_name_list = os.listdir(test_data_folder)
file_path_list = [
os.path.join(test_data_folder, x) for x in file_name_list
]
for file_name in file_name_list:
data_str = re.findall(r'([0-9\-]+)_', file_name)[0]
stock_name = re.findall(r'[0-9\-]+_([A-Za-z0-9]+)_', file_name)[0]
price_change = float(re.findall(r'#([0-9\.\-]+)#', file_name)[0])
baseline_dict[data_str][stock_name]['actual'] = price_change
test_date_set.add(data_str)
test_stock_set.add(stock_name)
# baseline algorithm
train_last_date_list = train_data_list[-NUMBER_OF_PREVIOUS_WEEK:]
complete_test_date_list = train_last_date_list + sorted(
list(test_date_set))
for last_date in train_last_date_list:
for test_stock in list(test_stock_set):
baseline_dict[last_date][test_stock][
'actual'] = train_baseline_dict[last_date][test_stock][
'actual']
#
#test_date_set.update(set(train_last_date_list))
#
for test_date in sorted(list(test_date_set)):
this_week_index = complete_test_date_list.index(test_date)
p1_week_index = this_week_index - 1
p2_week_index = this_week_index - 2
p1_week_date = complete_test_date_list[p1_week_index]
p2_week_date = complete_test_date_list[p2_week_index]
for stock in list(test_stock_set):
p1_pc = baseline_dict[p1_week_date][stock]['actual']
p2_pc = baseline_dict[p2_week_date][stock]['actual']
#predicted_price = (p1_pc + p2_pc) / 2 # baseline algorithm1
predicted_price = p1_pc # baseline algorithm2
baseline_dict[test_date][stock]['predict'] = predicted_price
print("complete_test_date_list: ", complete_test_date_list)
print("test_data_set: ", test_date_set)
print("test_stock_set: ", test_stock_set)
#print ("baseline_dict.dates: ", baseline_dict.items())
#
# get the predicted value
if is_random:
predict_pc_list = []
for i, test_date in enumerate(sorted(list(test_date_set))):
if random_seed:
random.seed(i + random_seed)
best_stock = random.sample(test_stock_set, 1)[0]
actual_pc = baseline_dict[test_date][best_stock]['actual']
predict_pc_list.append(actual_pc)
print("predict_pc_list: ", predict_pc_list)
elif is_highest_profit:
predict_pc_list = []
for i, test_date in enumerate(sorted(list(test_date_set))):
highest_return = float('-inf')
for stock in test_stock_set:
stock_return = baseline_dict[test_date][stock]['actual']
if stock_return > highest_return:
highest_return = stock_return
predict_pc_list.append(highest_return)
print("predict_pc_list: ", predict_pc_list)
else:
predict_stock_list = []
predict_pc_list = []
rmse_list = []
for test_date in sorted(list(test_date_set)):
complete_predict_value_list = []
complete_actual_value_list = []
highest_pc = float('-inf')
for stock in list(test_stock_set):
predict_value = baseline_dict[test_date][stock]['predict']
actual_value = baseline_dict[test_date][stock]['actual']
complete_predict_value_list.append(predict_value)
complete_actual_value_list.append(actual_value)
if predict_value > highest_pc:
highest_pc = predict_value
best_stock = stock
rmse = calculate_rmse(complete_actual_value_list,
complete_predict_value_list)
rmse_list.append(rmse)
predict_stock_list.append(best_stock)
actual_pc = baseline_dict[test_date][best_stock]['actual']
predict_pc_list.append(actual_pc)
print("predict_pc_list: ", predict_pc_list)
print("avg_pc: ", np.average(predict_pc_list))
print("avg_rmse: ", np.average(rmse_list))
return predict_pc_list
# ==========================================================================================================
parent_folder = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
mlp_path = os.path.join(parent_folder, 'general_functions')
sys.path.append(mlp_path)
# ==========================================================================================================
# ==========================================================================================================
# local package import
# ==========================================================================================================
from trade_general_funcs import calculate_rmse
# ==========================================================================================================
# ==========================================================================================================
# TEST-[1] <<<RMSE>>>
# ==========================================================================================================
# TODO complete the unit test
y_actual = np.array([1, 2, 3.2, 8.7, 5.5, 112])
y_predicted = np.array([2, 3.9, 4.1, 0, 34.3, 9.9])
rms1 = math.sqrt(mean_squared_error(y_actual, y_predicted))
rms2 = calculate_rmse(y_actual, y_predicted)
print("rms1: {}".format(rms1))
print("rms2: {}".format(rms2))
# ==========================================================================================================
# <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
# IMPORT IMPORT IMPORT IMPORT IMPORT IMPORT IMPORT IMPORT IMPORT IMPORT IMPORT IMPORT IMPORT IMPORT IMPORT I
# >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>