Bart van Merriënboer
Bart van Merriënboer
So I've been frustrated with the current state of logging in Blocks for a while. Pickled Python dictionaries have problematic memory requirements and aren't ideal for analysis given all the...
Not sure if this is a bug or not, but [this line](https://github.com/bartvm/blocks/blob/master/blocks/bricks/recurrent.py#L139) takes an `Application` instance instead of the new `BoundedApplication` instance. I think it should take the latter. @rizar...
We should use the new `equizip` function to make sure that when `as_dict=True` is used, the outputs of an application don't get cut off because they were left unnamed. This...
See discussion at https://github.com/bartvm/blocks/pull/336#issuecomment-75925893
When adding the YouTube audio dataset we briefly discussed this issue: Long sequence data needs to be split up into subsequences. We could write a transformer for this (potentially sharing...
Introduced in #196. Should be a very simple fix. See https://github.com/mila-udem/fuel/pull/271#issuecomment-175602758.
Tests should use the `skip_if_not_available` function to be skipped if certain datasets aren't available. On a clean install without datasets the following tests currently fail because `mnist.hdf` or `cifar10.hdf` is...
It is often possible to determine from the iteration scheme and the dataset how many batches will be returned (e.g. `ceil(num_examples / batch_size)`). This information could be useful (e.g. for...
Just sent them an e-mail about it: > I recently ran into what seems to be a bug in Scrutinizer's static code analysis. Consider any recent analysis log of the...
@yaoli was interested in using a divide-and-conquer approach to preprocessing, as is used in in @dwf's ImageNet PR (https://github.com/bartvm/fuel/pull/68). With that code, I think it should be relatively easy to...