Pascal Lamblin

(I put the label "Data Organization" as "Performance Evaluation" is not there, not sure if it is on purpose).

No shuffling of examples in introduction notebook

2

We realized that in the [introduction notebook](https://github.com/google-research/meta-dataset/blob/master/Intro_to_Metadataset.ipynb), the usage examples given for the `make_multisource_episode_pipeline` did not set the `shuffle_buffer_size` parameter, which defaults to not shuffling examples within each class. Two...

Plan for wrapping Fuel datasets into Pylearn2

7

One of the goals for factoring Fuel out of Blocks was to be able to re-use it as a new dataset back-end in Pylearn2. Here is a proposal and a...

Segfault in hdf5 on linux 32 bit when calling fuel-convert

2

It seems to happen reliably on one machine with Ubuntu 32-bit, where hdf5 is installed by Anaconda. I tried all the available versions of hdf5/pytables/h5py available on Anaconda, all exhibit...

Scripts like `fuel-download` used to be outside of the `fuel` package, and not tested for coverage. Now, we need them to be importable, so #211 moved them to `fuel.bin`, and...

Slurm back-end

2

It would be nice to have a way of using smartdispatch as a front end for slurm clusters.

Currently, `platoon_launcher` expects two scripts, `_controller.py` and `_worker.py` to exist. If we want to have a few generic controllers, rather than one per experiment, it should be possible to explicitly...