如何将存储在3个数据帧中的3个机器学习模型的结果合并/集合化，并输出1个数据帧并获得大多数人同意的结果？

Question

I am currently participating in an online hackathon. All the top entries are within 1% of each other. So I decided to run 3 different models instead of a single best performing one, ie ensemble learning, tuned hyperparameters on each one of them and then combine results of all three to get a better model. I've combined results of all three in a dataframe, it's df.head() is as below:

index | building_id | rf_damage_grade | xg_damage_grade | lr_damage_grade   | damage_grade

0   a3380c4f75  Grade 4 Grade 2 Grade 3 Grade 4

1   a338a4e653  Grade 5 Grade 5 Grade 5 Grade 5

2   a338a4e6b7  Grade 5 Grade 5 Grade 5 Grade 5

3   a33a6eaa3a  Grade 3 Grade 2 Grade 4 Grade 3

4   a33b073ff6  Grade 5 Grade 5 Grade 5 Grade 5

So 'rf_damage_grade' is the column of my best classifier. It gives around 74% accuracy, other two give 68% and 58% respectively. In final output i want, if 'xg_damage_grade' and 'lr_damage_grade' both agree on one value the final output 'damage_grade' gets changed to that value, otherwise it remains equal to the output of 'rf_damage_grade'. There are more than 400k rows in the data and and every time I rerun my model it is taking around an hour to do this on my Early 2015 MBP. Following is the code i've written:

for i in range(len(final)):
    if final.iloc[i,2]==final.iloc[i,3]:
        final.iloc[i,4]=final.iloc[i,2]
        if final.iloc[i,3]!=final.iloc[i,1]:
            count+=1
    else:
        continue

What can I do to make it more efficient? Is there any inbuilt function in sklearn to do this sort of thing?

Answer 1

Simply run conditional logic with .loc :

df.loc[df['xg_damage_grade'] == df['lr_damage_grade'], 'damage_grade'] = df['xg_damage_grade']
df.loc[df['xg_damage_grade'] != df['lr_damage_grade'], 'damage_grade'] = df['rf_damage_grade']

Or with numpy's where :

df['damage_grade'] = np.where(df['xg_damage_grade'] == df['lr_damage_grade'],
                              df['xg_damage_grade']
                              df['rf_damage_grade'])