double_lift.py

# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at https://mozilla.org/MPL/2.0/.

def double_lift(y_model1, y_model2, y_actual, model_key, model_type="additive", weight=None, p_tile=5, y_denom=None,
               rescale=False, y_name="Agg"):
    '''Creates a Double Lift Chart given 2 model prediction sets'''

    from functools import reduce
    import pandas as pd
    import seaborn as sns; sns.set()
    import matplotlib.pyplot as plt

    global df_DblLft
    global df_ChartAgg
    global mergelist
    if y_model1.name==y_model2.name:
        y_model1.rename(y_model1.name + "_model1", inplace=True)
        y_model1.rename(y_model1.name + "_model2", inplace=True)

    mergelist=[y_model1, y_model2, y_actual]
    if y_denom is not None:
        mergelist.append(y_denom)
    if weight is not None:
        mergelist.append(weight)


    df_DblLft=reduce(lambda left, right: pd.merge(left, right, on=model_key),  mergelist)

    if weight is None:
        df_DblLft["weight"]=1
        weight=df_DblLft["weight"]
    df_DblLft["ModelsRatio"]=y_model1/y_model2
    df_DblLft.sort_values(by="ModelsRatio", inplace=True)

    df_DblLft["runtot_weight"]=df_DblLft[weight.name].cumsum()

    df_DblLft["p_tile"]=pd.qcut(df_DblLft["runtot_weight"],p_tile, labels=False)

    ###Aggregate to display level

    if model_type=="additive":
        df_ChartAgg=df_DblLft.groupby("p_tile").agg(model1_agg=(y_model1.name, 'mean'),
                                                model2_agg=(y_model2.name, 'mean'),
                                                actual_agg=(y_actual.name, 'mean')
                                                )


    ###Handle Ratio Style
    if model_type=="ratio":
        df_DblLft["y_actual_num"]=df_DblLft[y_actual.name]*df_DblLft[y_denom.name]
        df_DblLft["y_model1_num"]=df_DblLft[y_model1.name]*df_DblLft[y_denom.name]
        df_DblLft["y_model2_num"]=df_DblLft[y_model2.name]*df_DblLft[y_denom.name]
        df_ChartAgg=df_DblLft.groupby("p_tile").agg(model1_num_agg=("y_model1_num", sum),
                                                model2_num_agg=("y_model2_num", sum),
                                                actual_num_agg=("y_actual_num", sum),
                                                y_denom_agg=(y_denom.name, sum)
                                                )

    df_ChartAgg["model1_agg"]=df_ChartAgg.model1_num_agg/df_ChartAgg.y_denom_agg
    df_ChartAgg["model2_agg"]=df_ChartAgg.model2_num_agg/df_ChartAgg.y_denom_agg
    df_ChartAgg["actual_agg"]=df_ChartAgg.actual_num_agg/df_ChartAgg.y_denom_agg


    df_ChartAgg=df_ChartAgg.filter(items=["actual_agg", "model1_agg", "model2_agg", "p_tile"])

    ###Handle Rescaling
    if rescale==True:
        df_ChartAgg.model1_agg= df_ChartAgg.model1_agg/df_ChartAgg.actual_agg
        df_ChartAgg.model2_agg= df_ChartAgg.model2_agg/df_ChartAgg.actual_agg
        df_ChartAgg.actual_agg=1


    ### Create Title Names
    TitleDict={5:"Quintile", 10:"Decile", 25:"Quartile"}
    if p_tile in TitleDict.keys():
        Title= TitleDict[p_tile]
    else:
        Title=p_tile + "%'iles"
    if y_name==None: y_name=model_type

    df_ChartAgg.reset_index(inplace=True)
    if y_name!="Agg":
        for colname in df_ChartAgg.columns:
            df_ChartAgg.rename(columns={colname:colname.replace('_agg','_'+y_name)}, inplace=True)

    FinalDf=df_ChartAgg.melt("p_tile",var_name='Model',value_name=y_name)
    FinalDf["p_tile"]=FinalDf["p_tile"]+1
    plt.xlabel(Title)

    DblLftCht= sns.catplot(x="p_tile",y=y_name,hue="Model",data=FinalDf,kind='point')
    DblLftCht.set_xlabels(Title)
    plt.title("Double Lift Chart")

    return DblLftCht