utils.py 文件源码

def rank_cat(df_tr,ycol,df_te=None,cols=None,rank=True,tag=''):
    if cols is None:
        cols = [i for i in df_tr.columns.values if df_tr[i].dtype=='object']
    if len(cols)==0:
        print("no cat cols found")
        return
    for col in cols:
        dic = df_tr.groupby(col)[ycol].mean().to_dict()
        if rank:
            ks = [i for i in dic]
            vs = np.array([dic[i] for i in ks]).argsort().argsort()
            dic = {i:j for i,j in zip(ks,vs)}
        df_tr[tag+col] = df_tr[col].apply(lambda x: dic[x])
        if df_te is not None:
            df_te[tag+col] = df_te[col].apply(lambda x: dic.get(x,np.nan))

#overfitting! try LOO!