Raw CNN & FAM model generation
Also you need Keras, which can be installed easily from pip.
It is important that .keras/keras.json, looks something like this:
{
"image_dim_ordering": "th",
"epsilon": 1e-07,
"floatx": "float32",
"backend": "tensorflow"
}
For fam_generate.py be sure to install the gr-specest fork from:
https://github.com/chrisruk/gr-specest
doxygen models.dox
Please use the FAM model at the moment, I'm just tweaking the CNN model, as I need to implement 64 sample shifts.
I'm currently investigating an issue I've found with the Keras implementations, where there is seemingly an 8% reduction in accuracy to my TFlearn implementation with RadioML data for the raw CNN model, however I believe this reduction could apply to both models.