My workflow tends to be time-series predictive ana...
# datasciencej
jimn
10/07/2019, 9:53 AMMy workflow tends to be time-series predictive analysis. Obtain database access, dump all of it to a local source, resample the frequencies, perform narrowing group-by, create the one-hot encodings, perform minmax normalization, and train/test models in Keras. The resample stage requires explicit parallelization in pandas typically via Apache arrow. One hot encodings may inflates the in-mem footprint to hundreds of gigabytes for initial scoping. Keras is explicitly hardware optimized for training and testing tensorflow data.