search

kristpapadopoulos / seriesnet

Time series prediction using dilated causal convolutional neural nets (temporal CNN)
MIT License
233 stars 64 forks source link

Test example #3

Closed saubersf closed 6 years ago

saubersf commented 6 years ago

Can you please provide an example of how to run the code? I tried: a = np.array([0,1,2,3,4,5]) evaluate_timeseries(a , 1) but it doesn't work.

kristpapadopoulos commented 6 years ago

This is correct way to run it. This should run - there was an issue with Keras merge layer that has now been replaced with Add but this timeseries may not produce meaningful results even it runs.

Each layer in the Dilated Convolution layer block has a dilated rate that looks back in time. The number of layers plus the dilation rate need to be adjusted to optimize for shorter length time series.

In the paper - the baseline code with seemed to work well on time series 108 or 64 time steps with 7 layers up to dilation rate of 64.