search

albertotb / solar

Solar data from OAHU
4 stars 0 forks source link

Train two 1D recurrent CNNs #11

Open roinaveiro opened 5 years ago

roinaveiro commented 5 years ago

One for latitude the other for longitude.

albertotb commented 5 years ago

Starting with Victor's code I'm doing both 1D convolutions in the notebook train_model-Conv_LSTM.ipynb

albertotb commented 5 years ago

Not trivial to add convolution to LSTM. There is a Keras layer for 2D conv, ConvLSTM2D that maybe can also be used for 1D (kernel 1xk and input 1xm). Right now we use TimeDistributed to apply the same kernel to every time step