from keras.models import model_from_yaml
import numpy as np
import argparse
parser = argparse.ArgumentParser()
parser.add_argument("--which_epoch", default=200, type=int)
parser.add_argument("--what_encoder", default=0, type=int)
args = parser.parse_args()
#load data
dataset_obj = Emilya_Dataset(window_width=200,
shift_step=20,
sampling_interval=None,
with_velocity=False,
number=None,
nb_valid=None,
nb_test=None)
test_X = dataset_obj.test_X[:, :, 1:]
test_Y2 = dataset_obj.test_Y2[:]
test_Y2 = convert_indices_2_onehot(test_Y2, nb_labels=8)
del dataset_obj
##load encoder
which_epoch = args.which_epoch
encoder_name = '../encoder' + str(which_epoch) + '.yaml'
with open(encoder_name, mode='r') as fl:
encoder = model_from_yaml(fl)
encoder.load_weights(encoder_name[:-4] + 'h5')
if args.what_encoder == 0:
import sys
import os
root_path = os.getenv('Seq_AAE_V1')
sys.path.append(root_path)
from Seq_AAE_V1.datasets.dataset import Emilya_Dataset
from Seq_AAE_V1.models.Seq_AAE.seq_aae_new_loss import Sequence_Adversrial_Autoencoder_with_New_Loss
dataset_obj = Emilya_Dataset(window_width=200,
shift_step=20,
sampling_interval=None,
with_velocity=False)
model = Sequence_Adversrial_Autoencoder_with_New_Loss(
latent_dim=50,
latent_activation='tanh',
latent_BN=False,
hidden_dim_enc_list=[100, 100],
activation_enc_list=['tanh', 'tanh'],
hidden_dim_dec_list=None,
activation_dec_list=None,
hidden_dim_dis_list=[100, 40],
activation_dis_list=['relu', 'relu'],
dropout_dis_list=[0.0, 0.0],
batch_size=200,
max_epoch=401,
optimiser_autoencoder='rmsprop',
optimiser_dis='Adam',
lr_autoencoder=0.001,
lr_dis=0.001,
decay_autoencoder=0.0,
decay_dis=0.0,
