I'm having trouble training an RNN and LSTM in Keras (Tensorflow backend). I'm using the NASA C-MAPSS turbofan engine data. My RNN performs horrible, much worse than an MLP, when there are published examples of RNNs doing *much* better than an MLP.
This may require you to reshape the data as required by Keras. I expect a MSE of less than 3000, more like somewhere around 2000.
We have an experience of 4+ years in machine learning and 3+ years in deep learning using sklearn, keras and tensorflow. Also Implemented RCNN , CNN and LTSM in keras and have very relevant problem solving experience using different kaggle data sets.