Plot using real time instead of time steps

TensorFlow General Discussion
,
1

Hi,
I am training a time series model to predict a time series curve using other parameters; however, I do not want to use the date as a feature. This means I take the datetime column out of my training input and then perform the batching, shuffling, training, etc. Is there a way that I can then now plot the model predictions using the real time values rather than the ambiguous time step values? Basically, it would be nice that the time was attached to the label time series values without actually being used in the training/prediction process. Is there a way to do this? I essentially followed this tutorial Time series forecasting  |  TensorFlow Core.

Thanks!

3

Hi @JMManos,

Sorry for the delay in response.
Yes, you can train your model without using the datetime column, but keep it separately. After predicting, match the predicted values with their corresponding datetime. Combine the datetime, actual, and predicted values into a table and then plot the predictions with the real datetime on the x-axis for a clear visualization. Here’s a sample implementation gist for your reference.

Hope this helps.Thank You.