Improvement suggestion for "Timeseries forecasting for weather prediction"

[To3No7]

As a teacher in ML I find there are clear problem spots in timeseries_weather_forecasting.py

First, the Data Preprocessing is not sufficient, compare this code to the TensorFlow tutorial on Time series forecasting, where they take care of Wind velocity at -9999, and also feature engineers the wind direction as well as the time parameter. This preprocessing is important to get good results.

In the TensorFlow tutorial, they also split the data into training, validation, and test sets, while this Keras example only do a split into training and validation sets, which is not a very good practice.

I also think this example would benefit from a baseline (like just predicting the same as the last sample/temperature) and then comparing the LSTM model’s performance to this. In connection with this, one can also show how to extract the test input and target vectors from a datastream by looping over the batched data.

[divyashreepathihalli]

@PrabhanshuAttri can you please take a look at this issue? Thank you!

[Zekrom-7780]

@sachinprasadhs @divyashreepathihalli , can I work on this Issue?

[github-actions[bot]]

This issue is stale because it has been open for 180 days with no activity. It will be closed if no further activity occurs. Thank you.