search

dfalbel / speech-keras

Code for the blog post "Simple Audio Classification with Keras"
https://blogs.rstudio.com/tensorflow/posts/2018-06-06-simple-audio-classification-keras/
11 stars 14 forks source link

Unable to replicate results #4

Closed AugustT closed 5 years ago

AugustT commented 5 years ago

Running the Markdown document I am unable to replicate the success reported, instead my classifier is not better than chance.

This is a real head-scratcher. I am running the document exactly as here. When i get to run the fit_generator my accuracy stays at ~0.03-0.04. I have tried a fresh installation of keras and anaconda, I have double checked the df is correct, but I cannot work out why we get such different results using the same code.

I'm really lost on this one so any thoughts/suggestions much appreciated. Running on windows... in case that would make a difference?!

dfalbel commented 5 years ago

@AugustT I will rerun the code this week. It could be related to tf python version.

dfalbel commented 5 years ago

FWIW I re-runed the code today with tf-1.13.1 and got the same accuracy as before. Could be some problem when downloading the data?