num_classes.txt not found

[lu3do]

I'm trying to run this (using docker) and ran into some small issues but was able to fix those myself.
for example tables module was missing, so I added that to the Dockerfile (will submit a PR for it later)
RUN pip --no-cache-dir install sqlalchemy pyyaml spotipy tables

But now i'm running into the next issue which is the missing num_classes.txt

I wanted to use the pretrained model, so copied that to the data/out but when running the classify it throws this error:

File "/notebooks/genres/train.py", line 26, in <module>
    with open('%s/%s' % (out_dir, 'num_classes.txt')) as f:
IOError: [Errno 2] No such file or directory: 'data/out/num_classes.txt'

Not sure what that file should contain? or how it can be generated.

[lu3do]

same goes for the num_features.txt file btw,

I guess it needs the configuration values for tensorflow, but want to know which values you've used.

[dpraul]

Oh wow, didn't anticipate that this repo would ever see any interest, but glad to help as much as I can. It's been a while since I've touched this project (and last I did, it was rushed, so there are definitely some poor design choices in here -- bear with me).

num_classes.txt and num_features.txt are generated by the graph step (for some reason), seen here: https://github.com/dpraul/genres-are-cool/blob/master/genres/graph.py#L95. Each file is just a single integer to help the neural network. Classes = genres, features = Spotify sound features. Certainly not the best way to do it, but hindsight.

I believe there are 15 genres, so num_classes.txt should just hold the number 15. There are 13 Spotify features, so num_features.txt should hold the number 13.

Let me know if that helps you along! Sadly, I just last week deleted all my local files for this repo, or else I'd be able to hand off more to you.

[dpraul]

In regards to your Docker issue, I image that stems from my poor foresight in not locking the Dockerfile to a particular version of Tensorflow, which has had many updates since I created this repo. Checking the GCR for Tensorflow, I imagine I was using version 0.9.0, linked here: https://console.cloud.google.com/gcr/images/tensorflow/GLOBAL/tensorflow@sha256:1e4ccc1a590cd52c6bfda6ee9f6ae7c2da8ef80e67e744532d1d5010cd92e9b2/details/info. You might try modifying the Dockerfile to use that version

[dpraul]

Ah, just kidding! I have a link to the docs for Tensorflow 0.10.0 in the README - use that version.

[lu3do]

Ok will try with that specific version.

Yea it also amazes me that this repo doesn't have a lot of activity.
I'm trying out ML for the first time, I want to build something to match a song to playlist(s) based on a song's audio features against the audio features of all songs already in a playlists. So I think your repo is a good starting point.

[dpraul]

Sounds fun! Best of luck. This repo was my first stab at ML, so do keep in mind that I'm sure there are amateurish and incorrect parts throughout, but I hope it serves useful as a reference