Black-Friday | winning entry for the Online Hackathon | Frontend Framework library

You replaced missing values in the train dataset, but not on the test one. So your prediction on the test is lower than the number of rows of your test dataset.

You should apply your operations (like replacing NA's) on the full dataset, and split in train/test after. This way, your both datasets will be similar, and prediction will better work.

There are several ways to transform this variable. In the picture I see, that there are not only bins, but also value '55+', it needs to be considered.

1) One liner:

As mentioned here by @MartijnPieters, etree's .findall uses the namespaces argument while the .register_namespace() is used for xml output of the tree. Therefore, consider mapping the default namespace with an explicit prefix. Below uses doc but can even be cosmin.

Additionally, consider with and enumerate() even the csv module as better handlers for your print and CSV outputs.

If I am getting all that right, than the answer is: No, not possible, those variables are local and the only way to make them global is, well, to use !global, to make them global. I another Option:

A @function get-variables($primary, $accent) which returns a map containing all the variables as key, value pairs, which can be called where ever needed.

Or you need to change your code in a way that @mixin generate-theme itself includes all other mixins with appropriate parameters.