# gradient clipping - to avoid gradient exploding
GRADIENT_CLIPPING = 2.5
scaler = StandardScaler()
autoencoder_path = "/home/suroot/Documents/train/daytrader/models/autoencoder-" + str(
encoding_dim) + ".hdf5"
cache = "/home/suroot/Documents/train/daytrader/autoencoded-" + str(
encoding_dim) + ".npy"
savePath = r'/home/suroot/Documents/train/daytrader/'
path = r'/home/suroot/Documents/train/daytrader/ema-crossover' # path to data
model_name = "raw_ts_model2"
use_cache = True
if (not use_cache or not os.path.isfile(cache)):
data = dt.loadData(path)
for i in range(data.shape[0]):
data[i, ] = (data[i, ] / data[i, -20]) - 1.0
# scale data .. don't forget to stor the scaler weights as we will need them after.
data = scaler.fit_transform(data)
print(data.shape)
# cache this data and the scaler weights.
np.save(cache, data)
# TODO: cache the scaler weights
print("loading cached data")
full_data = np.load(cache)
holdout = full_data[0:holdout_size, :]
data = full_data[holdout_size:, :]
print(data.shape)
num_stacked_layers = 2
# gradient clipping - to avoid gradient exploding
GRADIENT_CLIPPING = 2.5
scaler = StandardScaler()
cache = "/home/suroot/Documents/train/daytrader/seq2seq_raw-"+str(encoding_dim)+".npy"
#savePath = r'/home/suroot/Documents/train/daytrader/'
#path =r'/home/suroot/Documents/train/daytrader/ema-crossover' # path to data
savePath = r'/home/suroot/Documents/train/raw/'
path =r'/home/suroot/Documents/train/raw/22222c82-59d1-4c56-a661-3e8afa594e9a' # path to data
model_name = "pca2_ts_model"
use_cache = False
if( not use_cache or not os.path.isfile(cache) ):
data = dt.loadData(path, symbols=dt.CA_EXTRA)
#for i in range(data.shape[0]):
# data[i,] = (data[i,]/data[i,-20]) - 1.0
# scale data .. don't forget to stor the scaler weights as we will need them after.
data_scaled = scaler.fit_transform(data)
pca = PCA(n_components=encoding_dim, svd_solver='full')
data_reduced = pca.fit_transform(data_scaled)
print(data.shape)
# cache this data and the scaler weights.
np.save(cache, data_reduced)
# TODO: cache the scaler weights
print("loading cached data")
full_data = np.load(cache)
holdout = full_data[0:holdout_size,:]
data = full_data[holdout_size:,:]
from sklearn.neighbors import NearestNeighbors
from sklearn.decomposition import PCA
import matplotlib.pyplot as plt
import seaborn as sns
sns.set(color_codes=True)
plt.rcParams['interactive'] == True
# fix random seed for reproducibility
np.random.seed(90210)
subset = -1
path = r'/home/suroot/Documents/train/daytrader/ema-crossover' # path to data
data = dt.loadData(path, subset)
(data, labels) = dt.centerAroundEntry(data, -20)
print(data.shape)
print(np.sort(labels))
print("min: " + str(labels.min()))
print("max: " + str(labels.max()))
sns.distplot(labels)
plt.show()
(data2, labels2) = dt.filterOutliers(data, labels, 0.018, -0.016)
sns.distplot(labels2)
plt.show()
from sklearn.neighbors import NearestNeighbors from sklearn.decomposition import PCA import matplotlib.pyplot as plt import seaborn as sns sns.set(color_codes=True) plt.rcParams['interactive'] == True # fix random seed for reproducibility np.random.seed(90210) subset = -1 path = r'/home/suroot/Documents/train/daytrader/ema-crossover' # path to data data_unsorted = dt.loadData(path, subset) data = np.array(sorted(data_unsorted, key=lambda x: (x[-20] / x[-1]) - 1.0)) print(data.shape) labels = data[:, -1] entries = data[:, -20] print((labels[0] / entries[0]) - 1.0) print((labels[len(labels) - 1] / entries[len(entries) - 1]) - 1.0) #dt.plotTrainingExample(data[0,:]) dt.plotTrainingExample(data[-1, :]) dt.plotTrainingExample(data[-2, :]) dt.plotTrainingExample(data[-3, :])
# fix random seed for reproducibility
np.random.seed(90210)
crop_future = -20
class_to_view = 5 # class
num_classes = 5
input_size = 128
#savePath = r'/home/suroot/Documents/train/daytrader/'
#path =r'/home/suroot/Documents/train/daytrader/ema-crossover' # path to data
savePath = r'/home/suroot/Documents/train/raw/'
path =r'/home/suroot/Documents/train/raw/22222c82-59d1-4c56-a661-3e8afa594e9a' # path to data
original = dt.loadData(path, symbols=dt.CA_EXTRA)
(data, labels_classed, _) = dt.cacheLoadData(path, crop_future, num_classes, input_size, symbols=dt.CA_EXTRA)
print(data.shape)
x_train, x_test, y_train, y_test = train_test_split(data, labels_classed, test_size=0.1)
_, x_original, _, _ = train_test_split(original, labels_classed, test_size=0.1)
print("TEST SIZE: " + str(x_test.shape))
#/home/suroot/Documents/train/raw/mlp[3]-28-0.48.hdf5
from keras.models import load_model
model = load_model(savePath+'mlp[5]-48-0.41.hdf5')
model.summary()
class5Data = []