pandas groupby分组一学就会

GroupBy

df.groupby(cols, dropna)
df.groupby(condition)

cols, str or list.
condition, series of group names, matching rows.
dropna is a bool value,
indicating whether regard NaN as a group.
Return a DataFrameGroupBy or SeriesGroupBy.

GroupBy[cols] return a GroupBy.

attributes

ngroups, num of groups.
groups, dict[group_name, index]. \

function:
size, size of groups.
get_group, get group in origin data.

Aggregate function

All return a new dataframe.

Normally:

1. max, min, idxmax, idxmin
2. mean, sum, prod, std, var, size, count, nunique
3. all, any
4. median, quantile
5. sem, standard error of the mean of groups.
6. skew, unbiased skew.

agg

gb.agg(list), list, a list of function name.
gb.agg(dict), dict, dict[col_name, List[func] | List[(name, func)]].

agg returns dataframe with origin col and func.__name__(name) as column.

Inner function input: column in group;
output: a aggregate value

Transform function

Transform the origin dataframe.

Normally, cumcount, cumsum, cumprod, cummax, cummin

transform

gb.transform(func)
func, input: a column in group; output: a column in group.
The function puts inner output together as output.

Filter function

gb.filter(func)
func, input: a column in group; output: a series of bool.
The function filters data according the inner output.

apply

gb.apply(func)
func, input: a dataframe in group; output: scalar | Series | DataFrame.
apply will automatically split joint the inner output.