Thank you Marco, I am glad you found my blog post useful. The challenge that you are experiencing is related to the Time Series Analysis. The motivation behind this choice and the data split lies in Financial Econometrics.

The training and test data split has been performed in such a way that for each day's price has been predicted using the last 90 days (in case of Bitcoin) price data. For example, the January 10 has been determined using the price data from 11/10/2020-09/01/2021 .

Hope this helps, good luck!