timeseries_traffic_forecasting.py

[afcostauw]

Someone can explain to me how to make inference using a trained model using the code in timeseries_traffic_forecasting.py?

Imagine that a have historical data from 45 timesteps for 33 roads, and that I want to predict the next 2 timesteps for each road, i. e., input_sequence_length = 3 and forecast_horizon=2.

I do not understand why the result of model.predict(historical_data) has shape (41,2,33).
I was expecting having as a result an array of size (2,33), i. e., an array with the predict values for the 2 wanted timesteps, for each road.

Anyone can help me?

[sampathweb]

Its probably the rows in your X dataset you are passing to the model.predict. Please create a sample Colab notebook if its not resolved.

[pg37013]

@sampathweb
Thanks for the answer, but they are not the rows in the X dataset I'm passing to the model.predict.

I'm using this exactly script.

The result of the model.predict doesn't have the values of the X dataset, but an approximation for each one of the rows, which does not make sense to me. Giving the X dataset to model.predict forecast 2 timesteps, I was expecting have as output something with only 2 rows, each one representing the forecast timesteps, but this is not happening and I can not figure out what the model.predict's output represents.

I can give you a sneak peak (with a small sample of shape (18, 33)) about what is happening. I'm giving this dataset to model.predict:

And the model is giving as result an array with shape (14, 2, 33). The following print represents the array (14, 1, 33):

And the following print represents the array (14, 2, 33):