Tf.reduce_sum, tf.reduce_mean

1

what will the equivalent code be for

 z_loss = 0.5 * tf.reduce_sum(tf.square(z_mean) + tf.exp(z_logvar) - z_logvar  - 1, axis = [1,2,3])

What are the pytorch equivalent for reduce_mean and reduce_sum

Thanks

2

torch.mean and torch.sum would be the replacements (or call .mean() or .sum() on a tensor directly).

3

@ptrblck

cosim = tf.maximum(cosine_similarity(tensor_a, tensor_b), 1e-1) ** cossim_pow

can we use torch.clap_min instead of tf.maximum ??for above tensor flow implementation in pytorch

cosim = torch.pow(torch.clamp_min(cosine_similarity(tensor_a, tensor_b), 1e-1), cossim_pow)

4

Yes, clamp_min can be used instead of maximum also in PyTorch. The difference is that torch.max would expect tensors while clamp_min accepts scalar values for its threshold.