def train(self):
# Unwrap the matrices (put the most recent data on the bottom)
pivot = self.row_count % self.matrix_length
data = self.X[pivot:, :self.num_features]
data = np.concatenate((data, self.X[:pivot, :self.num_features]), axis=0)
y = self.X[pivot:, self.num_features]
y = np.concatenate((y, self.X[:pivot, self.num_features]), axis=0)
if (self.init_training or runnable(data) > 0.5):
#self.w_opt, self.a_opt, self.b_opt, self.S_N = normalTrain(data, y)
self.w_opt, self.a_opt, self.b_opt, self.S_N = train(data, y)
self.init_training = True
# Log current training windows as pickle files
if self.using_backup:
with open(self.X_backup_file, 'wb') as outfile:
pickle.dump(self.X, outfile)
def train(self):
# Unwrap the matrices (put the most recent data on the bottom)
pivot = self.row_count % self.matrix_length
data = self.X[pivot:, :self.num_features]
data = np.concatenate((data, self.X[:pivot, :self.num_features]), axis=0)
y = self.X[pivot:, self.num_features]
y = np.concatenate((y, self.X[:pivot, self.num_features]), axis=0)
if (self.init_training or runnable(data) > 0.5):
#self.w_opt, self.a_opt, self.b_opt, self.S_N = normalTrain(data, y)
self.w_opt, self.a_opt, self.b_opt, self.S_N = train(data, y)
self.init_training = True
# Log current training windows as pickle files
if self.using_backup:
with open(self.X_backup_file, 'wb') as outfile:
pickle.dump(self.X, outfile)
'''
print "X_og: \n",X_og
print "X: \n",X[cur_row]
# Train the model
if (row_count % forecasting_interval == 0 and
(row_count >= matrix_length or init_training)):
# Unwrap the matrices (put the most recent data on the bottom)
data = X[cur_row:, :num_sensors]
data = np.concatenate((data, X[:cur_row, :num_sensors]), axis=0)
y = X[cur_row:, num_sensors]
y = np.concatenate((y, X[:cur_row, num_sensors]), axis=0)
# BLR train:
w_opt, a_opt, b_opt, S_N = train(data, y)
init_training = True
# Log current training windows as pickle files
with open(XLOG_FILENAME, 'w') as logfile:
pickle.dump(X, logfile)
with open(Xog_LOG_FILENAME, 'w') as logfile:
pickle.dump(X_og, logfile)
# Make a prediction
if init_training:
# Prediction is dot product of data and weights
x_n = X[(row_count) % matrix_length][:num_sensors]
print "w_opt,\n", w_opt
'''
print "X_og: \n", X_og
print "X: \n", X[cur_row]
# Train the model
if (row_count % forecasting_interval == 0
and (row_count >= matrix_length or init_training)):
# Unwrap the matrices (put the most recent data on the bottom)
data = X[cur_row:, :num_sensors]
data = np.concatenate((data, X[:cur_row, :num_sensors]), axis=0)
y = X[cur_row:, num_sensors]
y = np.concatenate((y, X[:cur_row, num_sensors]), axis=0)
# BLR train:
w_opt, a_opt, b_opt, S_N = train(data, y)
init_training = True
# Log current training windows as pickle files
with open(XLOG_FILENAME, 'w') as logfile:
pickle.dump(X, logfile)
with open(Xog_LOG_FILENAME, 'w') as logfile:
pickle.dump(X_og, logfile)
# Make a prediction
if init_training:
# Prediction is dot product of data and weights
x_n = X[(row_count) % matrix_length][:num_sensors]
print "w_opt,\n", w_opt
