search

metrofun / E3D-LSTM

Unofficial PyTorch implementation of E3D-LSTM
127 stars 33 forks source link

Understanding the hyperparameters? #6

Open IanQS opened 2 years ago

IanQS commented 2 years ago

Hey there!

Thanks for the library. I was wondering if I could get some insight into what these hyperparameters "are"? I've downloaded the dataset and played around with it setting breakpoints, but the names are confusing me.

        self.input_time_window = 4
        self.output_time_horizon = 1
        self.temporal_stride = 1
        self.temporal_frames = 2
        self.time_steps = (
            self.input_time_window - self.temporal_frames + 1
        ) // self.temporal_stride

input_time_window

I'm not sure what this is....

output_time_horizon

I'm guessing that output_time_horizon is the length of the prediction? E.g if output_time_horizon == 1, output a scalar, if output_time_horizon == 2 output a vector of length 2?

temporal stride

Given some input timeseries of [1,2,3,4,5,6, 7] with stride 2 and size 3

T1 = [1,2,3] T2 = [3,4,5]

?

temporal_frames

Is this the window size as mentioned above?