L2_loss = gluon.loss.L2Loss()
huber_loss = gluon.loss.HuberLoss()
initializer = mx.initializer.MSRAPrelu()
optimizer = 'adam'
optimizer_params = {'learning_rate': args.learning_rate}
trainer_params_list = {
'batch_size': args.batch_size,
'epoch_num': args.epochs,
'loss_func': abs_loss,
'initializer': initializer,
'optimizer': optimizer,
'optimizer_params': optimizer_params,
'patience': args.patience
}
train_mark = 'testing'
nn_trainer(train_mark,
model1,
test_sub_valid_X,
test_future_test_X,
test_sub_valid_Y,
sub_train_nd,
sub_valid_X,
future_test_X,
sub_valid_Y,
trainer_params_list=trainer_params_list,
ctx=ctx,
model_name=args.model_name)
batch_size = 128
n_epochs = 500
"""
The model training
"""
### The model parameters
abs_loss = gluon.loss.L1Loss()
L2_loss = gluon.loss.L2Loss()
huber_loss = gluon.loss.HuberLoss()
initializer = mx.initializer.MSRAPrelu()
optimizer = 'adam'
optimizer_params = {'learning_rate': 0.01}
trainer_params_list = {
'batch_size': batch_size,
'epoch_num': n_epochs,
'loss_func': huber_loss,
'initializer': initializer,
'optimizer': optimizer,
'optimizer_params': optimizer_params
}
train_mark = 'testing'
nn_trainer(train_mark,
model1,
sub_train_nd,
conv_valid_X,
sub_valid_X,
sub_valid_Y,
trainer_params_list=trainer_params_list,
ctx=ctx)
with open('feature_prepare.pkl', 'rb') as f:
[trainX_dt,trainX2_dt, trainY_dt,trainY2_dt, testX_dt, testX2_dt,testY_dt,testY2_dt] = pickle.load(f)
sub_train_X, sub_train_Y = nd.array(trainX_dt, ctx=ctx), nd.array(trainY_dt, ctx=ctx)
sub_valid_X, sub_valid_Y = nd.array(testX_dt, ctx=ctx), nd.array(testY_dt, ctx=ctx)
future_train_X, future_test_X = nd.array(trainY2_dt, ctx=ctx), nd.array(testY2_dt, ctx=ctx)
sub_train_nd = gluon.data.ArrayDataset(sub_train_X, future_train_X, sub_train_Y)
model1 = TCN()
#choose parameters
batch_size= 512
n_epochs=500
"""
The model training
"""
### The model parameters
abs_loss = gluon.loss.L1Loss()
L2_loss = gluon.loss.L2Loss()
huber_loss = gluon.loss.HuberLoss()
initializer = mx.initializer.MSRAPrelu()
optimizer = 'adam';
optimizer_params = {'learning_rate': 0.5}
trainer_params_list = {'batch_size': batch_size,'epoch_num':n_epochs,
'loss_func': abs_loss, 'initializer': initializer,
'optimizer':optimizer, 'optimizer_params':optimizer_params}
train_mark='testing'
nn_trainer(train_mark, model1, sub_train_nd, sub_valid_X,future_test_X, sub_valid_Y, trainer_params_list=trainer_params_list, ctx=ctx)
def optimize(optimizer_params):
"""
The model optimization
"""
# Keep track of evals
global ITERATION, out_file, epochs
ITERATION += 1
### The model parameters
abs_loss = gluon.loss.L1Loss()
L2_loss = gluon.loss.L2Loss()
huber_loss = gluon.loss.HuberLoss()
initializer = mx.initializer.MSRAPrelu()
optimizer = 'adam'
optimizer_params['units'] = int(optimizer_params['units'])
optimizer_params['batch_size'] = int(optimizer_params['batch_size'])
print("Iteration %d: Creating model" % ITERATION)
model1 = TCN(input_dimention=args.dim,
output_ax=args.horizon,
units=optimizer_params['units'],
dropout=optimizer_params['dropout'])
train_mark = 'optimizing'
#epochs = int(batch/batches[0])
trainer_params_list = {
'batch_size': optimizer_params['batch_size'],
'epoch_num': epochs,
'loss_func': huber_loss,
'initializer': initializer,
'optimizer': optimizer,
'optimizer_params': {
'learning_rate': optimizer_params['learning_rate']
},
'patience': args.patience,
'iteration': ITERATION,
'units': optimizer_params['units'],
'dropout': optimizer_params['dropout'],
'lr': optimizer_params['learning_rate']
}
print("Training model ... ")
st_time = datetime.now()
valid_loss, valid_ND, valid_NRMSE, rho10, rho50, rho90 = nn_trainer(
train_mark,
model1,
test_sub_valid_X,
test_future_test_X,
test_sub_valid_Y,
sub_train_nd,
sub_valid_X,
future_test_X,
sub_valid_Y,
trainer_params_list=trainer_params_list,
ctx=ctx,
model_name=args.model_name)
eval_time = str(datetime.now() - st_time)
print("Iteration %d: Getting results ... " % ITERATION)
return {
'loss': valid_ND,
'ND': valid_ND,
'NRMSE': valid_NRMSE,
'val_loss': valid_loss,
'params': optimizer_params,
'rho_metric': {
'rho10': rho10,
'rho50': rho50,
'rho90': rho90
},
'iteration': ITERATION,
'eval_time': eval_time,
'status': STATUS_OK
}