Missing Indicator | Random Sample Imputation | Handling Missing Data Part 4 

same ques as he already used fit transform on that x_train_trf2 and x_test_trf2 above why he is using again below then if using add_indicator =True)

There are two ways to do the same task. Mistakenly he mentioned X_train_Trf2 and X_test_trf2 at start which should be at end

from sklearn.compose import ColumnTransformer from sklearn.impute import SimpleImputer from sklearn.preprocessing import StandardScaler # Select the numerical columns numeric_cols = ['age', 'income'] # Create the preprocessing pipelines for the numerical columns numeric_pipeline = Pipeline([ ('imputer', SimpleImputer(strategy='random_sample')), ('scaler', StandardScaler()) ]) # Create the ColumnTransformer preprocessor = ColumnTransformer([ ('numeric', numeric_pipeline, numeric_cols) ])

Suppose you have 1 column on which you want to apply ordinal encoder and on any other column onehotencoder so here you have to first split those columns from the dataset for applying those two transformers on it and after applying again you have to merge those two columns with dataset.....so it seems very time consuming and boring, right ?....so here comes the superhero whose name is ColumnTransformer and with the help of column transformer you can apply both those transformers on both columns in a single code and that too without splitting them from the dataset which reduces our lot of efforts......so in the case where you want to apply more than one transformer on more than one column then simply use columntransformer. When to use pipeline: We have trained our ML model on the preprocessed data but when the model will be in the production then the users will not provide us the preprocessed data....so here we use pipeline in which the input by the user goes and pass through all the preprocessing steps we performed while training the model and in the pipeline we passes columntranformers because those are the heroes who took our dirty data and gave us the preprocessed data and here they do the same with the input from real time users and in pipeline we also passes our model. In piepline the output form one step acts as a input for the next step and after completion of all the steps, it goes to our model for prediction and that's why we pass model in pipeline. So in the production, we use the strength of both column transformer and pipeline together. I know kuch zyada hi explain kar diya par nalla baitha tha so I thought thoda revision hi karlu😮‍💨...I hope it would help you

@@Alive-Ness hahaha you did a great job in explaining the concept.....I got it..... thank you very much man🙏🙏

I also wanted to same question

len(X_train) is correct

Change float to int.

instead of int(df.col), use => df.col.astype(int)

Yup

observation is not a dataframe there

I am also getting the same error..........