deepest-debug-challenge's Introduction

Deepest Debug Challenge: can you squash them all?

WARNING: This model is implemented to NOT work correcly. DO NOT use this for your application. It will hurt you.

This is a quiz code for debugging a badly-implemented neural network.

The code runs fine and trains the model somehow, but not correctly.

You are given a task to fix all the bugs in the code.

This code is for Quest 2 of Deepest Season 5 Recruiting.

Updates

12/05 It seems like the original description in the link above was not clear enough.
Therefore, I rewrote some of the the descriptions just to make the requirements more clear.
Please refer to this link: http://bit.ly/2ANdfMW
12/12 If you fix all the bugs, only one epoch of training would be enough to get an accuracy of around 80%. Therefore, those who do not have any GPU, please feel free to tackle this problem. (In fact, I myself did not use any GPU till I finished designing this quest.)
12/19 I found an unnecessary function in data_utils.py and deleted it. This change would have no effect on the applicants and the challenge itself. (It was kind of a test code written in my Jupyter notebook, and I accidentally copy-and-pasted it while I was making all the code into python files. My bad :/)

Use python >= 3.6.

run python prepro.py to make data/prepro.npz, which is the data used to train and test the model.
run python train.py to train and test the model.

Recommend Projects