How to interpret training evaluation?

Hello everyone,
Im using an RNN with LSTM layer to make predictions on a stock, and I also implement the deeplearning4j UI while training. The charts look very bumpy, which I know the fact that I am predicting stocks will somewhat cause this, but it looks very different from the pictures on the deeplearning4j website. Here is what the UI dashboard looks like after a good amount of training: .

My question is what is causing the bumpiness, especially in the bottom two graphs? The top graph from what I have seen isn’t too bad. Also at the end of training I evaluate the model on the test data, and this is the result i get: . It doesn’t look too bad from what I can tell. But could anyone give their take on those stats too?