Skip to content

Guidelines for setting T and k in mixLSTM #1

Description

@Simply-Adi

Hello, I came across your interesting paper highlighting the potential of mixLSTM in handling time-varying relationships. I was trying to examine how well mixLSTM handles a one-time shift in the following data:

x1 = np.random.normal(5,1,1000)
x2 = np.random.normal(2,5,1000)
y_base = 10+(3*x1[:600]*x1[:600])+(5*x2[:600])
y_shifted = 10+(6.2*x1[600:]*x1[600:])+(5*x2[600:])
y = np.concatenate((y_base,y_shifted))

For this (X,y), I was experimenting with:
args= {'d':2,'hidden_size':300,'output_size':1,'batch_size':32,'T':30 and above}
k is 2 by default.
I tried various values of the above parameters but the model was not able to detect the shift. Could you pls. suggest guidelines on setting T and k?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions