Feather is a binary format for storing and exchanging tabular data. Its binary storage and compression produce smaller files that pandas can read and write faster than CSV or JSON. section 12.4
Use Feather when a pandas workflow repeatedly saves and reloads a large data frame. Write with to_feather and reload with read_feather; the resulting data remains a pandas data frame. section 12.4
Benchmarking a large data set
For a parking-violations data set, a benchmark wrote JSON in about 46 seconds, CSV in 114 seconds, and Feather in 8 seconds. Reading took about 470 seconds for JSON, 35 seconds for CSV, and 9 seconds for Feather. The Feather file was also the smallest of the three measured outputs. section 12.6
df.to_feather('mydata.feather')
Inspect the compact pandas operation used to write a data frame as a Feather file, then compare it with the corresponding read operation. section 12.4
