def main():
stock_sql = StockSQL()
if VERSION == 1:
model = load_dnn_model_v1(weight_path=WEIGHT_PATH)
ret = load_and_pred_data_v1(stock_sql, model)
elif VERSION == 2:
model = load_dnn_model_v2(weight_path=WEIGHT_PATH)
ret = load_and_pred_data_v2(stock_sql, model)
elif VERSION == 3:
model = load_dnn_model_v3(weight_path=WEIGHT_PATH)
ret = load_and_pred_data_v3(stock_sql, model)
elif VERSION == 4:
model = load_dnn_model_v4(weight_path=WEIGHT_PATH)
ret = load_and_pred_data_v4(stock_sql, model)
elif VERSION == 5:
model = load_dnn_model_v5(weight_path=WEIGHT_PATH)
ret = load_and_pred_data_v5(stock_sql, model)
else:
model = load_dnn_model_v6(weight_path=WEIGHT_PATH)
ret = load_and_pred_data_v6(stock_sql, model)
print("上傳資料中.....")
# 上傳預測結果
stock_sql.write_multi_line("INSERT INTO `AI__分析a` VALUES (%s,%s,%s,%s,%s)",
ret)
print("Finish")
# Create X Input 3
x_data_3 = np.array(values)[:, 13:16].astype(np.float)
x_data_3 = (x_data_3 - mean_dict_3[stock_id]) / std_dict_3[stock_id]
# Create X Input 4
x_data_4 = np.array(values)[:, 16:20].astype(np.float)
x_data_4 = (x_data_4 - mean_dict_4[stock_id]) / std_dict_4[stock_id]
# Create X Input 5
x_data_5 = np.array(values)[:, 20:].astype(np.float)
x_data_5 = (x_data_5 - mean_dict_5[stock_id]) / std_dict_5[stock_id]
# Predict
outputs = model.predict((x_data_1, x_data_2, x_data_3, x_data_4, x_data_5))
# Save predict result
dates = [str(v[0]) for v in values]
ret += package_outputs(stock_id, dates, outputs, version)
print('Preparing Stock: {}, Number: {}'.format(stock_id, len(values)))
return ret
if __name__ == '__main__':
from units.sql import StockSQL
# 連線SQL資料庫
stock_sql_test = StockSQL()
# 載入資料
load_data(stock_sql_test, load_dir='1723')
from datetime import datetime
from flask import Flask, render_template
from flask_bootstrap import Bootstrap
from flask_moment import Moment
from units.draw import *
from units.sql import StockSQL
from flask_wtf import FlaskForm
from wtforms import SubmitField, DateField, SelectMultipleField, IntegerField, StringField
from wtforms.validators import DataRequired
app = Flask(__name__)
app.config['SECRET_KEY'] = 'woodylin7158517'
bootstrap = Bootstrap(app)
moment = Moment(app)
stock_sql = StockSQL()
class StocksForm(FlaskForm):
start_date = DateField('開始時間', id='datepick', validators=[DataRequired()])
end_date = DateField('結束時間', validators=[DataRequired()])
stock_id = IntegerField('股票代號', validators=[DataRequired()])
submit = SubmitField('送出')
@app.errorhandler(404)
def page_not_found(e):
return render_template('404.html'), 404
@app.errorhandler(500)
def main():
# 初始話參數
if TEST_MODE:
today = 'test'
else:
dt = datetime.datetime.today()
today = "{}-{}-{}".format(dt.year, dt.month, dt.day)
model_dir = os.path.join(
'outputs', '{}-{}-version{}.h5'.format(today, TRAINING_DATA, VERSION))
log_dir = os.path.join(
'logs', '{}-{}-version{}'.format(today, TRAINING_DATA, VERSION))
if os.path.isdir(log_dir) or os.path.isfile(model_dir):
print("檔案已經存在")
exit()
# 連線SQL資料庫
stock_sql = StockSQL()
# 載入資料 & 創建Model
if VERSION == 1:
x_train, y_train, x_val, y_val = load_data_v1(stock_sql,
load_dir=TRAINING_DATA)
model = load_dnn_model_v1()
elif VERSION == 2:
x_train, y_train, x_val, y_val = load_data_v2(stock_sql,
load_dir=TRAINING_DATA)
model = load_dnn_model_v2()
elif VERSION == 3:
x_train, y_train, x_val, y_val = load_data_v3(stock_sql,
load_dir=TRAINING_DATA)
model = load_dnn_model_v3()
elif VERSION == 4:
x_train, y_train, x_val, y_val = load_data_v4(stock_sql,
load_dir=TRAINING_DATA)
model = load_dnn_model_v4()
elif VERSION == 5:
x_train, y_train, x_val, y_val = load_data_v5(stock_sql,
load_dir=TRAINING_DATA)
model = load_dnn_model_v5()
elif VERSION == 6:
x_train, y_train, x_val, y_val = load_data_v6(stock_sql,
load_dir=TRAINING_DATA)
model = load_dnn_model_v6()
else:
x_train, y_train, x_val, y_val = load_data_v7(stock_sql,
load_dir=TRAINING_DATA)
model = load_dnn_model_v7()
# Training
model.compile(
tf.keras.optimizers.Adam(0.001),
loss=binary_focal_loss,
metrics=[keras.metrics.BinaryAccuracy(),
OverKill(),
UnderKill()],
run_eagerly=True)
model_tcbk = keras.callbacks.TensorBoard(log_dir=log_dir)
model_mckp = SaveBastOverKillAddUnderKillModel(model_dir)
model_acbk = OverKillAddUnderKill(log_dir)
model.fit(x_train,
y_train,
batch_size=64,
epochs=50,
validation_data=(x_val, y_val),
callbacks=[model_tcbk, model_mckp, model_acbk])