b = train_cond[sindex]
print(a.shape)
print(b.shape)
a = np.reshape(a, (-1, 32, 130))
b = np.reshape(b, (-1, 32, 12))
print(a.shape)
print(b.shape)
#initialize model
model = VAE(130, 2048, 3, 12, 128, 128, 32)
model.eval()
dic = torch.load("vae/tr_chord.pt")
for name in list(dic.keys()):
dic[name.replace('module.', '')] = dic.pop(name)
model.load_state_dict(dic)
if torch.cuda.is_available():
model = model.cuda()
print(model)
a = torch.from_numpy(a).float()
b = torch.from_numpy(b).float()
res = model.encoder(a, b)
z1 = res[0].loc.detach().numpy()
z2 = res[1].loc.detach().numpy()
print(z1)
print(z1.shape)
print(z2)
print(z2.shape)
z1 = torch.from_numpy(z1).float()
z2 = torch.from_numpy(z2).float()
res = model.decoder(z1, z2, b)
from torchvision import transforms
from torchvision.utils import save_image
import numpy as np
from tqdm import trange
from sa.model import MnistClassifier
from vae.model import VAE
img_size = 28*28*1
torch.no_grad() # since nothing is trained here
### Prep (e.g. Load Models) ###
vae = VAE(img_size = 28*28, h_dim = 1600, z_dim = 400)
vae.load_state_dict(torch.load('./vae/models/MNIST_EnD.pth'))
vae.cuda()
classifier = MnistClassifier(img_size = img_size)
classifier.load_state_dict(torch.load('./sa/models/MNIST_conv_classifier.pth'))
classifier.eval()
classifier.cuda()
print("models loaded...")
test_dataset = torchvision.datasets.MNIST(root='./data', train=False, transform=transforms.ToTensor(), download=False)
test_data_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=1, shuffle=True)
print("Data loader ready...")
### GA Params ###
gen_num = 500
pop_size = 50
best_left = 20