def getTrainData_1(symbol, startDate, endDate, periods=14, remove_head=19):
dates = pd.date_range(startDate, endDate)
df = util.loadPriceData([symbol], dates)
util.fill_missing_values(df)
# skip periods day latest (14 days) that predict a price is new high or not
df_sliced = df.ix[0:len(df) - periods]
price_close = pd.DataFrame(df_sliced[symbol])
set = pd.DataFrame(df_sliced['SET'])
rsi = ind.rsi(price_close) / 100 # normalize
sr = ind.create_dataframe_SR(df_sliced, [symbol])
myRatio = ind.get_myRatio(price_close)
daily = ind.daily_returns(price_close) * 100 # normalize
_, _, macd = ind.average_convergence(price_close)
ohlcv = util.load_OHLCV(symbol, dates)
percent_KD = ind.percent_KD(ohlcv) / 100 # normalize
c2o = ind.daily_returns_2(ohlcv) * 100 # normalize
volume = util.loadVolumeData([symbol], dates)
#skip periods day latest
volume_sliced = volume.ix[0:len(volume) - periods]
assert len(volume_sliced) == len(df_sliced)
obv = ind.OBV(volume_sliced, df_sliced)
obv_rsi = ind.rsi(obv) / 100 # calcuate momentum with rsi and normalize
set_rsi = ind.rsi(set) / 100 # calcuate momentum with rsi and normalize
# Join data frame
# rename column
rsi.rename(columns={symbol: 'RSI'}, inplace=True)
#sr.rename(columns={symbol:'SR'},inplace=True)
myRatio.rename(columns={symbol: 'MY'}, inplace=True)
daily.rename(columns={symbol: 'DAILY'}, inplace=True)
macd.rename(columns={symbol: 'MACD'}, inplace=True)
obv_rsi.rename(columns={symbol: 'OBV_RSI'}, inplace=True)
set_rsi.rename(columns={'SET': 'SET_RSI'}, inplace=True)
Xtrain = pd.DataFrame(index=df_sliced.index)
Xtrain = Xtrain.join(rsi['RSI'])
Xtrain = Xtrain.join(percent_KD['%K'])
#Xtrain = Xtrain.join(sr['SR'])
Xtrain = Xtrain.join(myRatio['MY'])
Xtrain = Xtrain.join(daily['DAILY'])
Xtrain = Xtrain.join(macd['MACD'])
Xtrain = Xtrain.join(c2o)
Xtrain = Xtrain.join(obv_rsi['OBV_RSI'])
Xtrain = Xtrain.join(set_rsi)
upTrend = ind.isUpTrend(df, symbol, periods=periods)
Ydigit = 1 * upTrend[
symbol] # multiply 1 : True is converted to 1 (up trend) , False becomes 0
# skip at head row, avoid NaN values
Xtrain = Xtrain.ix[remove_head:]
Ydigit = Ydigit.ix[remove_head:]
price_close = price_close.ix[remove_head:]
return Xtrain, Ydigit, price_close
def writeDataSetY(df, symbols, predict=0, window=20, output_dir=output_dataset): symbols = df.columns.values df_sr = ind.create_dataframe_SR(df, symbols) df_result = ind.isStrongSR(df_sr) # By defult, Use data of 20 days (backward) to predict 15 days (foward) # Because I would like to know sharpe ratio that are more than 0 continue 10 days or not shift = window + predict -1 df_result = df_result.shift(-1*shift) df_result=df_result.dropna(0) df_result.to_csv(join(output_dir, "@Dataset_Y.csv"), index_label='<DTYYYYMMDD>')
def max_sharpe_ratio(periods=5):
df_sr = ind.create_dataframe_SR(df, symbolList, window=periods)
df_sr.fillna(0, inplace=True) # fill NaN to 0
df_sr = df_sr.shift(-(periods - 1))
df_sr.dropna(0, inplace=True) # drop NaN at tail
result = np.mean(df_sr)
# Returns the indices that would sort an array
# max values at last element
maxIndexResult = np.argsort(result)
nameSymbol = [
df_sr.columns[index] for index in reversed(maxIndexResult)
if result[index] > 0
]
# save to dump file
pickle.dump(nameSymbol, open("symbol_list.p", "wb"))
print("List max sharpe ratio\n", nameSymbol)
df_norm = utl.normalize_data(df)
utg.plotStock(df_norm, nameSymbol[0:5], startDate, endDate,
'High sharpe ratio')
