Tailoring the load.py for using my own data

[kapilagrawal95]

Hi
I have a time series data almost similar to jsb_chorales data. The only difference is that for each timestep jsb_chorales has 88 binary values while I have one continuous value so essentially my one sequence can be represented in a 1-d array. I would like to train a dmm on my data.

I tried to understand the simulateLinearData but I couldn't entirely comprehend the code; mainly because of this line:
zlist= [np.copy(z)[:,None,:]]
I am unable to visualize it.

[rahulk90]

I could be misunderstanding your question but you don't need to use simulateLinearData for setting up the dataset. Yo have to format your data into a 3D tensor with shape: N x T x 1 (in your case since you have one-dimensional data)

You may find this ipython notebook helpful for setting up the data (See A] Data):
Notebook - Setup

[kapilagrawal95]

Thanks! That resolved my above query.
Another thing: my dataset's time-series vary. Will the algorithm work in that case? And will the 3-d tensor have shape: N x (T_max) x 1?

[rahulk90]

Yes. Search for mask in the ipython notebook. It should tell you how to specify variable length sequences.