Instructions

1. Unzip data from phase 1 and 2 as folders on top level. Name those folders "data" and "data2".
2. Run embed.ipynb
3. Run train.ipynb. If you wish to see the validation loss/accuracy as well, run with final=False near the top. Change test_number in the last cell if necessary. It is for naming folders to store tensorboard files and trained models.
4. Run test.ipynb. Change test_num to 1 or 2 depending on phase.
5. Remember to name the output files carefully, to not overwrite them.
6. Result may not be exactly the same, due to random factors and change of parameters I was unable to recover. Still, it should give somewhere around 72% at least.