Full article can be found here

The author tries to explain the reasons why the design of dataframes and tables are different.

database folks think in terms of tables (aka relations)
datascience folks think in terms of dataframes (read pandas)
former uses lists, whereas the latter uses sets
- author argues that this can have impact on how the implementations will differ between these cases use-cases!
- because lists are dependent on the order whereas sets aren't
matrices vs tables
- matrices are good at aggregation kind of operations
- however, by storing variables also as columns, such a "matrix" could be converted into a relation, thereby easing this process too!
however, tables seem to outshine dataframes when it comes to joins
- closest operation in dataframes is a merge
- author argues that merge and concat ops in dataframes are not actually the same as as in join and union ops in tables and these dataframes ops are needlessly complicated!
dataframes are procedural interfaces while relations offer more declarative style
sql ops, for eg, only support first-order logic (looping across items is 2nd order)
code reuse is better with dataframes than with tables
debugging is also better with dataframes
tables have query-optimizers which can lead to better optimized operations internally
- in dataframes, users have to manually tune for perf
dataframes offer many ways to solve one thing, which can be confusing to programmers