I have accomplished to get reproducible results for my Keras project following the tips in this stackoverflow question: Reproducible results using Keras with TensorFlow backend
However when restarting the kernel my Keras model (variational autoencoder/decoder) I still do get different results each time i restart the kernel.
To fix it i have already tried to;
- Manually set the weights and biases in the entire model(rather than random assignment).
- Set the PYTHONASHEED in the VENV to zero.
- Explored different versions from Keras and python3.
Any ideas are welcome because i'm stuck in this project.