Creating chunks of segments

[arilwan]

Hi,

Thank you for putting these immense resources in one place.

1. May I please ask why the creation of fixed-sized chunks/blocks of the training series?

2. Support I choose to use a classical machine learning classifier, such as random forest, would you also recommend creating these fixed-sized chunks first?

Regards.

[guillaume-chevalier]

Hi @arilwan,

Here are quick answers:

1. The RNN is not dynamic and uses fixed-size windows to make things simpler. Lenghts could have been varied at train time. Moreover the dataset was pre-chunked into these fixed-size windows, so doing otherwise wasn't considered here. Dynamic RNNs exists, but are more complicated to code (and to train).
2. I recommend using fixed-size chunks for other classifiers provided you've featurized it well, as done here. However, it happened to me to use non-fixed size windows, with RNNs, and without RNNs (using other classical machine learning algorithms such as a random forest). It all depends on what you are trying to solve and what data you have.

Regards,
Guillaume