Can't reproduce results

[kurianbenoy]

• On running the same code, the results are not reproducible

Please clarify if this is the exact code, or have I missed something

• Your models look pretty basic, it's hard to believe you can predict with so less data and simple model

[Ascalonic]

Dense LSTM networks are usually tuned to perform predictions on batches of data much larger than this. We barely have a test batch. But time series prediction is something RNNs and LSTMs are pretty good at. In our data, it outperformed fbProphet very well. At the end of the day, what works, works. And this seemed to work better than https://www.corona.megstalks.com

One out of 4-5 runs yields results like this. There is comparatively high noise in the data and LSTM networks seem to amplify them in some test runs. To get the correct results, try running them until the mean squared error and loss are acceptable. For instance, I just ran it again in Colab and got:

Train Score: 34.21 RMSE, Loss: 3.8427e-05 and a perfectly curved prediction line.

• What was the RMSE in your case?