aa9e1d66eed11e425d177ad6af114877b736bb25,rbm/rbm_complex.py,RBM,train,#RBM#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#,314

Before Change

                                      r=lr)

        vis = self.generate_visible_space()
        for ep in progress_bar(range(epochs + 1), desc="Epochs ",
                               total=epochs, disable=disable_progbar):
            
            random_permutation = torch.randperm(data.shape[0])

            shuffled_data           = data[random_permutation]   
            shuffled_character_data = character_data[random_permutation]

            batches = [shuffled_data[batch_start:(batch_start + batch_size)] 
                       for batch_start in range(0, len(data), batch_size)]

            char_batches = [shuffled_character_data[batch_start:(batch_start + batch_size)] 
                            for batch_start in range(0, len(data), batch_size)]

            if ep % log_every == 0:
                logZ = self.log_partition(vis)
                nll = self.nll(data, logZ)
                tqdm.write("{}: {}".format(ep, nll.item() / len(data)))

            if ep == epochs:
                break

            stddev = torch.tensor(
                [initial_gaussian_noise / ((1 + ep) ** gamma)],
                dtype=torch.double, device=self.device).sqrt()

            for batch in progress_bar(batches, desc="Batches",
                                      leave=False, disable=True):

                grads = self.compute_batch_gradients(k, batch, char_batches,
                                                     l1_reg, l2_reg,

After Change

        vis = self.generate_visible_space()
        print ("Generated visible space. Ready to begin training.")
        fidelity_list = []
        epoch_list = []

        for ep in range(epochs+1):
        
            random_permutation = torch.randperm(data.shape[0])

            shuffled_data           = data[random_permutation]   
            shuffled_character_data = character_data[random_permutation]

            batches = [shuffled_data[batch_start:(batch_start + batch_size)] 
                       for batch_start in range(0, len(data), batch_size)]

            char_batches = [shuffled_character_data[batch_start:(batch_start + batch_size)] 
                            for batch_start in range(0, len(data), batch_size)]

            if ep % log_every == 0:
                //logZ = self.log_partition(vis)
                //nll = self.nll(data, logZ)
                fidelity_ = self.fidelity(vis, "Z" "Z")
                print ("Epoch = ",ep,"\nFidelity = ",fidelity_)
                fidelity_list.append(fidelity_)
                //print("Not calculating anything right now, just checking grads.")

            if ep == epochs:
                fidelity_file = open("fidelity_file.txt", "w")
                print ("Finished training. Saving results..." )               
                for i in range(len(fidelity_list)):
                    fidelity_file.write("%.5f" % fidelity_list[i] + " %d\n" % epoch_list[i])
                break

            stddev = torch.tensor(
                [initial_gaussian_noise / ((1 + ep) ** gamma)],
                dtype=torch.double, device=self.device).sqrt()

            for batch_index in range(len(batches)):

                grads = self.compute_batch_gradients(k, batches[batch_index], char_batches[batch_index],
                                                     l1_reg, l2_reg,
                                                     stddev=stddev)

Instances