Prediction with Pylearn2

[gngdb]

Need a function that will run Pylearn2 for prediction based on supplied run settings. Should be in test.py.

[gngdb]

There may be some hint for how this could be done in Pylearn2's own predict scripts.

Also, there will be a problem with decoding the prediction numbers; as they are encoded into integers in the Dataset class we've written and I don't know if this is something that's saved in the pickle file (I'm guessing it's not). Won't be clear how to fix this problem until we know more about how the prediction script should be written.

[gngdb]

To make this work there are a few things to do. First part is to produce a submission csv with integer labels (ie in the encoding that the model already knows):

• [x] Update initialisation of dense_dataset to allow test set loading, while keeping track of image labels.
• [x] Get predictions out of Pylearn2 in the same order
• [x] Write these to csv

Then, once that works we can start worrying about matching the labels to the integer labels the model already knows; ie the problem described above.

• [x] Decode integer labels

[gngdb]

Fixed it so the order of the labels the model will know will be correct. However, note that any models trained on the old code will be using the wrong map!