code:
Notes
the code now takes an argument that must be train or predict
I have checked in the file ds_model_k_lasing.h5, this is saved model that did well.
log file of a training run for that model: log_of_training_ex03.output
...
- What is it training on? Is it really lasing, or some systematic bias in run?
- Predict on different lasing run - some confidence
- Predict on different no-lasing files
- With batch normalization, during training, we normalize over batch with no-lasing and lasing
- Good to test predictions on batch with only one or the other, test the saved moving average of the layers
- Keras has option to normalize over batches during prediction also, could help regularize
Exercises
predict on a mix of lasing no lasing, just a lasing