...
- Keras using theano
- pre-procecess images with log transofrm and grand mean subtraction
- classifying the two color shot - 4 outputs, just the lasing runs
...
- The train function can be modified to pass 'test' to the getTrainData function. Otherwise reads all data (takes about a minute)
- The separate getTrainData function can be used from interactive ipython
- large batch size of 64, will use too much memory on interactive nodes, reduce for testing
- new function TFModel.build_2color_model
- We will print the total loss, which includes the regularization term, as well as the cross entropy
- For guided backprop, just print class 3 - both colors lased
Exercises
- compare guided backprop to unguided - just a straight derivative of image