| import torch | |
| import numpy as np | |
| from machinedesign.autoencoder.interface import load | |
| from keras.models import Model | |
| torch.use_deterministic_algorithms(True) | |
| model = torch.load("mnist_deepconvae/model.th") | |
| model_keras = load("/home/mehdi/work/code/out_of_class/ae/mnist") | |
| print(model_keras.layers[8]) | |
| m = Model(model_keras.inputs, model_keras.layers[8].output) | |
| X = torch.rand(1,1,28,28) | |
| with torch.no_grad(): | |
| # X1 = model.sparsify(model.encode(X)) | |
| X1 = model(X) | |
| X2 = model_keras.predict(X) | |
| X2 = torch.from_numpy(X2) | |
| print(torch.abs(X1-X2).sum()) | |
| # for i in range(128): | |
| # print(i, torch.abs(X1[0,i]-X2[0,i]).sum()) | |
| # print(X1[0,i, 0, :]) | |
| # print(X2[0,i,0, :]) | |