Reduces RAM utilization by manipulating data types properly towards low size data types (from Int64>>Int8,Int16,Int32 and from Float64 to Float32 or Float 16.)
Source: https://www.kaggle.com/gemartin/load-data-reduce-memory-usage
Note: If you have problems with this function try to discard Float16 part, as there might be errors because of this data type, look: pandas-dev/pandas#9220