Sean Robertson

Batches are independent - to process more in parallel and also to mitigate effects of outliers in the dataset ([batch normalization](https://gab41.lab41.org/batch-normalization-what-the-hey-d480039a9e3b)). They don't relate to order of inputs.

You can remove references to sconce and `job`, it's a plotting library (not davidlmorton/pytorch-sconce) that is not production ready.

More leftover code... in this case the tutorial [was using](https://github.com/spro/practical-pytorch/blob/0cc55f5aaed44e7903edb8842671411301fcf003/seq2seq-translation/seq2seq-translation.ipynb) a different strategy (Ctrl-F "location-based" in [Effective Approaches to Attention-based Neural Machine Translation](https://arxiv.org/abs/1508.04025)). In that case the attention is calculated...

Right & right. The max length is also used to filter training data.

There are many more iterations than training examples, so it is fairly likely to cover them all. A more reliable way would be to go through the examples in order...

👍 That's the one. As mentioned in the readme: https://github.com/spro/practical-pytorch/tree/master/char-rnn-generation

The `u''` is just to show they are Unicode strings.

Looks good, thank you. If you could clean up some of the commented out lines I'll merge it in. By the way this would be great to have in https://github.com/spro/char-rnn.pytorch

Make sure your dataset is exactly the same, because `lineToTensor` relies on the number and order of characters in `all_characters` to create the input tensors. Another solution is to make...

Good idea thanks, I'll add this in the next round of updates.