durner / storn-keras Goto Github PK

I am a little confused by 'evaluate_online' methd in STORN.py: Why do we need a ground_truth when doing online anomaly detection, and what should I put into it? I mean, since the output(or prediction) of the STORNModel is kind of distribution parameters, it seems reasonable to detect next step anomaly from the loss which could be got from 'keras_variational' function in variational.py.

Hello,
I am wondering to know the type and shape of inputs for STORNModel. I simply input (n_sample, seq_len, dim) data to 'inputs' and 'target', but I got errors. It looks the inputs should be a list of some numpy arrays. Its slightly difficult to analyze what the desired type and shape of inputs from the code. Could you tell me what can be the structure of inputs and target in fit function?

Ex)
m = STORNModel()
m.fit(inputs, target)

THanks!

Great job!
I'm very interested in the project and I'd like to make it run on my PC, but I don't have a Baxter robot, and can not generate the time series data. Could you please kindly upload a copy of the data? Thanks!