YouTube video review: Multivariate time series forecasting using LSTM by Digitalsreeni

One great thing about Python’s Tensorflow library is the fact that it will perform a lot of different machine learning tasks. One machine learning task that Tensorflow is capable of performing is that of time series analysis and forecasting. In order to acquire skill in this area, I have been watching YouTube videos on the subject. One such video is 181 — Multivariate time series forecasting using LSTM by DigitalSreeni.

This was a good video to watch and the content creator was kind enough to include the code in a Github repository, so I only needed to copy and paste the code into Google Colab, the Jupyter Notebook that I normally use.

One thing the content creator did was to use sklearn’s StandardScaler() function to scale and then descale the data later. This concerns me because Tensorflow has normalisation functions that can easily be used, so I am not quite sure why he chose to add an extra nuance to the code.

When I tried to run the code up there were errors in the final two lines, which I had to correct. I suspect this is because there have been updates to the libraries, which has affected the operationality of the code. This is one reason why it is important to stay abreast of changes to libraries and amend the code as appropriate because programs may not work otherwise.

I spent a day studying the code that the content creator made and hopefully I learned a little bit more about how to use Tensorflow to make predictions on multivariate time series data.

The Github repository where the code for this video is stored can be found here:- https://github.com/bnsreenu/python_for_microscopists/blob/master/181_multivariate_timeseries_LSTM_GE.py