Bob Adolf

Currently, there's only a weak need for OpenCV in DeepQ, since we don't actively visualize the emulator's behavior. Given the minor headaches associated with installing it and the rather hefty...

The latter is the far more common use case, and it's easy to shoot ourselves in the foot otherwise.

We need to make sure everything falls under an MIT-compatible license before final release. Add a master LICENSE file which comprehensively assigns credit where it is due.

#13 helped, but the output is still a little messy to use. Tracking a loss curve is a very useful thing, but in practice, loss isn't sufficient a lot of...

Should be conditionally added to computational graph at model construction time.

Documentation

1

We need a set of docs describing both simple logistical issues as well as some intuition about the models' behavior and implementation.

Create small debug dataset

1

We can't expect to distribute real datasets. First, they're hundreds of GB, so it's a logistical headache. Second, we don't have the license approval to do so, even if we...

1. Configure Travis (or a private CI) to support all the libraries we have. 2. Write tests which use the debug datasets 3. Green badges = happy programmers

The ops required for training (loss function, optimizer, and training loop) are spread out over too many files. They are difficult to read and follow. These should be integrated into...

Datasets and models should be decoupled, both for general code hygiene and to allow other datasets to be plugged in later (see #4). A dataset object should contain methods to...