Pandas/python function for determining the fractions of examples falling into different groups, taking into account other grouping criteria.

group_fractions_pandas.py

def groupCountFractionals(dataframe,target,outer):
    '''
    dataframe: a pandas dataframe
    target: a string corresponding to the column of interest in the dataframe
    outer: a list of the columns by which the counts should be conditioned
    
    Returns the fraction of target_criteria_group / outer_criteria_group counts.
    
    Be mindful to take group sizes (Outer Count) into consideration. 
    As outer count gets smaller, the fraction value
    will obviously become increasingly large.
    
    '''
    dataframe["Count"] = 1
    outcome_variable = outer[0]
    inner_group_criteria = outer + [target]
    outer_counts = dataframe.groupby(outer).count()[["Count"]]
    outer_counts.columns = ["OuterCount"]
    outer_counts = outer_counts.reset_index()
    inner_counts = dataframe.groupby(inner_group_criteria).count()[["Count"]]
    inner_counts.columns = ["InnerCount"]
    inner_counts = inner_counts.reset_index()
    in_and_out =pandas.merge(inner_counts,outer_counts,on=outer,how="right")
    in_and_out["Fraction"] = in_and_out["InnerCount"] / in_and_out["OuterCount"]
    
    '''The following code is for tracking groups that have counts of zero'''
    group_values = [list(dataframe[o].unique()) for o in outer]
    inner_values = list(dataframe[target].unique())
    group_values.append(inner_values)
    group_permutations = list(itertools.product(*group_values))
    null_df = pandas.DataFrame(group_permutations)
    null_columns = outer.copy()
    null_columns.append(target)
    null_df.columns = null_columns
    
    null_df = pandas.merge(null_df,outer_counts, on = outer,how="outer")
    null_df["InnerCount"] = 0
    null_df["Fraction"] = 0
    
    absent_rows = pandas.merge(null_df,in_and_out,how="outer",on=null_columns,indicator="left_only")
    zero_rows = null_df[absent_rows["left_only"] == "left_only"]
    return pandas.concat([in_and_out,zero_rows]).fillna(0)