[constantcrying]

This is an insane article, I do not think the author has any idea what is going on.

The comparison of numpy reading CSV to arrow reading parquet is completely bizarre and totally misses the point of switching out the underlying data format.

[kbrkbr]

Or maybe you did not read it properly?

> Reading in that CSV file into memory would take Python 55.8 seconds, but PyArrow did the work in 11.8 seconds.

It's later clarified that Pyarrow does load csv here, though the numbers don't fully add up. Also the format change is explained.