def main():
# load keras model
kmodel = load_model('./models/cnn_model_10bar_ohlc.h5')
# load data
data = pro.load_pkl('./data/label8_eurusd_10bar_1500_500_val200_gaf.pkl')
train_x = data['train_ohlc_gaf']
train_label = data['train_label_onehot']
# create foolbox model
fmodel = KerasModel(kmodel, bounds=(-1, 1))
# create our modified attack model
MODIFIED_LocalSearchAttack = foolbox.attacks.LocalSearchAttack(model=fmodel)
# generate fake data
generate_adversarial_examples(kmodel = kmodel, fmodel = fmodel, attacker = MODIFIED_LocalSearchAttack,
x_data = train_x, y_label = train_label)
def main(params):
# load data
data = pro.load_pkl(params['data'])
# load our model
kmodel = load_model(params['model_name'])
# create foolbox model
fmodel = KerasModel(kmodel, bounds=(-1, 1))
# customized LocalSearchAttack
attack = foolbox.attacks.LocalSearchAttack(model=fmodel)
# attack all samples
results = attack_all_samples(data, kmodel, fmodel, attack)
return results
# loop training
result_dict = {
'train_result_cm': [],
'test_result_cm': [],
'total_acc': []
}
for i in range(PARAMS['n_loop']):
model, hist = train_model(PARAMS)
# collect testing results
collect_result(PARAMS, model, hist, result_dict)
# save all results to pickle file
filename = './results/merged_results_%s.pkl' % PARAMS['n_loop']
with open(filename, 'wb') as f:
pickle.dump(result_dict, f)
if __name__ == "__main__":
PARAMS = {}
PARAMS['data'] = pro.load_pkl('./data/merged_examples.pkl')
PARAMS['classes'] = 9
PARAMS['lr'] = 0.01
PARAMS['epochs'] = 1000
PARAMS['batch_size'] = 64
PARAMS['patience'] = 80
PARAMS['zero_weight'] = 2
PARAMS['optimizer'] = optimizers.SGD(lr=PARAMS['lr'])
PARAMS['n_loop'] = 100
PARAMS['model_path'] = './models/mergedata_model.h5'
main(PARAMS)
'''
plt.close()
fig = plt.figure(figsize=(4, 4))
ax = plt.subplot2grid((1, 1), (0, 0))
mpf.candlestick_ohlc(ax, ts_data, width=0.4, alpha=1,
colordown='#53c156', colorup='#ff1717')
plt.title(predict_result)
plt.show()
plt.close()
if __name__ == "__main__" :
PATTERN_LS = ["no_class", "evening", "morning", "bearish_engulfing", "bullish_engulfing",
"shooting_star", "inverted_hammer", "bearish_harami", "bullish_harami"]
PKL_NAME = "./pickle/label8_eurusd_1500_500_val200_gaf_culr.pkl"
MODEL_PATH= './model/checkpoint_model.h5'
data = pro.load_pkl(PKL_NAME)
model = load_model(MODEL_PATH)
_idx = int(args.i)
data_gaf = data['test_gaf'][_idx, :, :, :]
data_gaf = data_gaf.reshape(1, 32, 32, 4)
ts_data = data['test_data']
ts_data = np.c_[range(ts_data[_idx, :, :].shape[0]), ts_data[_idx, :, :]]
predict_result = evaluate(data_gaf, model)
result_picture(ts_data, predict_result)
candle_low = gen_imgs[:, :, 2]
candle_close = gen_imgs[:, :, 3]
max_in_o_c = np.maximum(candle_open, candle_close)
min_in_o_c = np.minimum(candle_open, candle_close)
gen_imgs[:, :, 1] = np.maximum(candle_high, max_in_o_c)
gen_imgs[:, :, 2] = np.minimum(candle_low, min_in_o_c)
util_plt.plot_vae_result(gen_imgs, filename)
def save_model(self):
self.decoder.save('./models/cvae_decoder.h5')
self.encoder.save('./models/cvae_encoder.h5')
if __name__ == '__main__':
PARAMS = dict()
PARAMS['number_of_bars'] = 10
PARAMS['epochs'] = 10001 # 10000001
PARAMS['latent'] = 100
PARAMS['batch_size'] = 100
PARAMS['sample_interval'] = 10000
PARAMS['data'] = util_pro.load_pkl(
'./data/label8_eurusd_10bar_1500_500_val200_gaf.pkl')
cvae = CVAE(PARAMS)
cvae.train(epochs=PARAMS['epochs'],
batch_size=PARAMS['batch_size'],
sample_interval=PARAMS['sample_interval'])
cvae.save_model()
result_dict = {
'train_result_cm': [],
'test_result_cm': [],
'total_acc': []
}
for i in range(params['n_loop']):
model, hist = train_model(params)
# collect testing results
collect_result(params, model, hist, result_dict)
# save all results to pickle file
filename = './results/clean_results_%s.pkl' % params['n_loop']
with open(filename, 'wb') as f:
pickle.dump(result_dict, f)
if __name__ == "__main__":
PARAMS = {}
PARAMS['data'] = pro.load_pkl(
'./data/label8_eurusd_10bar_1500_500_val200_gaf_culr.pkl')
PARAMS['classes'] = 9
PARAMS['lr'] = 0.01
PARAMS['epochs'] = 1000
PARAMS['batch_size'] = 64
PARAMS['patience'] = 80
PARAMS['zero_weight'] = 2
PARAMS['optimizer'] = optimizers.SGD(lr=PARAMS['lr'])
PARAMS['n_loop'] = 100
PARAMS['model_path'] = './models/orgdata_model.h5'
main(PARAMS)