Fitting column wise ordinal encoder

Question

I already posted this here but no response, so posting it here

I have a dataframe like as shown below

tdf = pd.DataFrame({'grade': np.random.choice(list('AAAD'),size=(5)),
                   'dash': np.random.choice(list('PPPS'),size=(5)),
                   'dumeel': np.random.choice(list('QWRR'),size=(5)),
                   'dumma': np.random.choice((1234),size=(5)),
                   'target': np.random.choice([0,1],size=(5))
})

I would like to get a mapping dictionary based on ordinal encoding technique as given here

from feature_engine.encoding import OrdinalEncoder
X = tdf.drop(['target'], axis=1)
y = tdf.target
train_t, test_t, y_train, y_test = train_test_split(X, y, 
                                                test_size=0.25,
                                                random_state=0)
cat_list= tdf.select_dtypes(include=['object']).columns.tolist()
ordinal_encoders = {}
for col in cat_list:
    print(col)
    ordi = OrdinalEncoder(encoding_method='ordered')
    ordinal_encoders[col] = ordi
    ordi.fit(train_t[col], y_train)
    train_t[col] = ordi.transform(train_t[col])

However, I get the below error

TypeError: X is not a pandas dataframe. The dataset should be a pandas dataframe.

How can I fit and transform the ordinal encoder on a column by column basis? I am able to get the encoders initialized as shown below but unable to fit and transform them

{'grade': OrdinalEncoder(),
 'dash': OrdinalEncoder(),
 'dumeel': OrdinalEncoder()}

I would like to do it this way because later, I wish to finally get the mapping dictionary (ordinal value for each of the catgories and store it in a dictionary)

Answer 1

that is because train_t[col] in the last two lines return a pandas Series object rather than a pandas DataFrame. Use train_t.loc[:, [col]] instead.

all in all, to fix your issue, replace the last for loop with:

for col in cat_list:
    print(col)
    ordi = OrdinalEncoder(encoding_method='ordered')
    ordinal_encoders[col] = ordi
    ordi.fit(train_t.loc[:, [col]], y_train)
    train_t[col] = ordi.transform(train_t.loc[:, [col]])