def get_predictions_input(config_filename, adjClose, datearray):
params = get_params(config_filename)
first_history_index = params['first_history_index']
num_periods_history = params['num_periods_history']
increments = params['increments']
print(" ... generating examples ...")
Xpredict, Ypredict, dates_predict, companies_predict = generateExamples3layerGen(datearray,
adjClose,
first_history_index,
num_periods_history,
increments,
output_incr='monthly')
print(" ... examples generated ...")
return Xpredict, Ypredict, dates_predict, companies_predict
def one_model_prediction(imodel, first_history_index, datearray, adjClose, symbols, num_stocks, verbose=False):
# --------------------------------------------------
# build DL model
# --------------------------------------------------
config_filename = imodel.replace('.hdf','.txt')
print("\n ... config_filename = ", config_filename)
#print(".", end='')
model = build_model(config_filename)
# collect meta data for weighting ensemble_symbols
params = get_params(config_filename)
#num_stocks = params['num_stocks']
num_periods_history = params['num_periods_history']
increments = params['increments']
symbols_predict = symbols
Xpredict, Ypredict, dates_predict, companies_predict = generateExamples3layerGen(datearray,
adjClose,
first_history_index,
num_periods_history,
increments,
output_incr='monthly')
dates_predict = np.array(dates_predict)
companies_predict = np.array(companies_predict)
# --------------------------------------------------
# make predictions monthly for backtesting
# - there might be some bias since entire preiod
# has data used for training
# --------------------------------------------------
try:
model.load_weights(imodel)
except:
pass
dates_predict = np.array(dates_predict)
companies_predict = np.array(companies_predict)
inum_stocks = num_stocks
cumu_system = [10000.0]
cumu_BH = [10000.0]
plotdates = [dates_predict[0]]
_forecast_mean = []
_forecast_median = []
_forecast_stdev = []
for i, idate in enumerate(dates_predict[1:]):
if idate != dates[-1] and companies_predict[i] < companies_predict[i-1]:
# show predictions for (single) last date
_Xtrain = Xpredict[dates_predict == idate]
_dates = np.array(dates_predict[dates_predict == idate])
_companies = np.array(companies_predict[dates_predict == idate])
#print("forecast shape = ", model.predict(_Xtrain).shape)
_forecast = model.predict(_Xtrain)[:, 0]
_symbols = np.array(symbols_predict)
indices = _forecast.argsort()
sorted_forecast = _forecast[indices]
sorted_symbols = _symbols[indices]
try:
_Ytrain = Ypredict[dates_predict == idate]
sorted_Ytrain = _Ytrain[indices]
BH_gain = sorted_Ytrain.mean()
except:
BH_gain = 0.0
avg_gain = sorted_Ytrain[-inum_stocks:].mean()
_forecast_mean.append(_forecast.mean())
_forecast_median.append(np.median(_forecast))
_forecast_stdev.append(_forecast.std())
if verbose:
print(" ... date, system_gain, B&H_gain = ",
idate,
format(avg_gain, '3.1%'), format(BH_gain, '3.1%'),
sorted_symbols[-inum_stocks:])
cumu_system.append(cumu_system[-1] * (1.+avg_gain))
cumu_BH.append(cumu_BH[-1] * (1.+BH_gain))
plotdates.append(idate)
print(" ... system, B&H = ", format(cumu_system[-1], '10,.0f'), format(cumu_BH[-1], '10,.0f'))
return cumu_system, cumu_BH, sorted_symbols, plotdates
