minimally sufficient pandas

minimally_sufficient_pandas.md

the tl;dr of https://medium.com/dunder-data/minimally-sufficient-pandas-a8e67f2a2428

select a column of data, use brackets df['column_name']

select rows of data, use .loc or datetime index df['2019-01-01':'2019-02-28' ]

• if performance is primary concern, using numpy array instead of Pandas

use read_csv and it's many arguments for reading files

use .isna method to filter NaN rows

use ~ to negate

prefer math operators (+ - * / ** // %) instead of math methods (lt gt eq ne)

use pandas math aggregation methods instead of built in math functions

• df['column_name'].sum() instead of sum(df['column_name'])
• df['column_name'].max() instead of max(df['column_name'])

prefer df.groupby(...).agg(...) for doing group by aggregation

• Good: df.groupby('grouping column').agg({'aggregating column': 'aggregating function'})
  • e.g. df.groupby('fruit').agg({'tastiness': 'mean'})
  • e.g. df.groupby('fruit').agg({'tastiness': 'mean', 'weight': ['mean', 'median']})
• OK: df.groupby('grouping column')['aggregating column'].agg('aggregating function')
  • e.g. df.groupby('fruit')['tastiness'].mean()

for going from wide to long format, prefer melt over stack

for going from long to wide format, prefer pivot_table over unstack or pivot