I have a dataset with historical sales of products within some chains. It looks like: date | customer | store | warehouse | product | volume, units 2000-01-01 customer1 store1 warehouse1 product1 10 2000-01-02 customer1 store1 warehouse1 product1 6 ... I want to do preprocessing with the dataset and work with each key (unique combination of a customer, a store, a warehouse, and a product) as a separate dataset because it takes some computational time to filter the key, preprocess it and put it back into the dataset. Putting the keys into separate datasets might solve the problem, but as I have a large number of keys (~200K), it doesn't look like the right way. I thought about dask, but I haven't found the way to set manually how to define buckets by keys. So, I'd like to address to each key like a separate dataset, but don't want to just save them into separate files. Any ideas? Continue reading...
