def main():
locations = [i for i in range(12,25)]
stacked_errors = []
for location in locations:
df1 = remove_trend(location, 27)
df2 = remove_trend(location, 26)
df3 = remove_trend(location, 28)
X1, y1, removed_seasonality1, removed_std1 = generate_lags(
df1, 20, 2 * 24)
X2, y2, removed_seasonality2, removed_std2 = generate_lags(
df2, 20, 2 * 24)
X3, y3, removed_seasonality3, removed_std3 = generate_lags(
df3, 20, 2 * 24)
tf.reset_default_graph()
merged_model = Merged_Model()
errors = merged_model.run_epochs(X1, X2, X3, y1, y2, y3, removed_seasonality1,
removed_seasonality2, removed_seasonality3,
removed_std1, removed_std2, removed_std3,
location)
stacked_errors.append(errors)
pd.DataFrame(stacked_errors, columns=["LGA", "JFK", "EWR"]).to_csv(
'data/multi_errors.csv', index=False, sep="\t")
def main():
hyper_config = HyperParam_Ranges()
fn = Auxiliary_funcs(hyper_config)
df = remove_trend(16, 27)
X_train, X_validation = fn.split_data_into_training_validation(
df['time series'].values)
for i in range(50):
internal_hidden_size, batch_size, learning_rate = fn.generate_random_hyperparams(
)
# construct hyperparam string for each combination -> Tensorboard
hparam = fn.make_hparam_string(internal_hidden_size, batch_size,
learning_rate)
# clear the default graph
tf.reset_default_graph()
config = Config(internal_hidden_size, batch_size, learning_rate)
model = RNN_NeuralModel(config)
model.run_epochs(X_train, X_validation, hparam)
def main():
loc_fixed = 23
df1 = remove_trend(loc_fixed, 24)
df2 = remove_trend(loc_fixed, 21)
df3 = remove_trend(loc_fixed, 22)
X1, y1, removed_seasonality1, removed_std1 = generate_lags(df1, 20, 2 * 24)
X2, y2, removed_seasonality2, removed_std2 = generate_lags(df2, 20, 2 * 24)
X3, y3, removed_seasonality3, removed_std3 = generate_lags(df3, 20, 2 * 24)
merged_model = Merged_Model()
merged_model.run_epochs(X1, X2, X3, y1, y2, y3, removed_seasonality1,
removed_seasonality2, removed_seasonality3,
removed_std1, removed_std2, removed_std3)
def main():
df = remove_trend(16, 28)
config = Config()
model = RNN_NeuralModel(config)
model.run_epochs(df['time series'].values,
df['removed seasonality'].values,
df['removed std'].values)
def main():
locations = [i for i in range(12,25)]
stacked_errors = []
for location in locations:
df = remove_trend(location, 27)
tf.reset_default_graph()
config = Config()
model = RNN_NeuralModel(config)
error = model.run_epochs(df['time series'].values,
df['removed seasonality'].values,
df['removed std'].values,
location)
stacked_errors.append(error)
pd.DataFrame(stacked_errors).to_csv(
'data/LGA_single_errors.csv', index=False, sep="\t")
with tf.name_scope("Cost") as scope:
mse = tf.reduce_mean(tf.pow(tf.subtract(pred, target_placeholder), 2.0))
with tf.name_scope("Optimize") as scope:
optimizer = tf.train.RMSPropOptimizer(learning_rate)
opt = optimizer.minimize(mse)
""" Run graph """
with tf.Session() as sess:
rmses = []
for dest in [26, 27, 28]:
df = remove_trend(16,dest)
X = df['time series'].values
removed_seasonality = df['removed seasonality'].values
removed_std = df['removed std'].values
stacked_preds = []
stacked_ground_truth = []
for i in range(bootstrap_size, len(X) , n_test):
print("Current window", i, i + n_test)
sess.run(tf.global_variables_initializer())
X_train = X[:i]
def main():
df = remove_trend(16, 27)
X, y, removed_seasonality, removed_std = generate_lags(df, 20, 2 * 24)
run_epochs(X, y, removed_seasonality, removed_std)