LabelEncoder won't encode strings?

Question

I'm trying to encode the non-numeric columns of a pandas df to numeric values. I'm using

df = df.fillna('0')
msk = np.random.rand(len(df)) < 0.8
df_train = df[msk]
df_test = df[~msk]
columns_to_encode = df.select_dtypes(exclude=[np.number]).columns
encoder_dict = {col: LabelEncoder() for col in columns_to_encode }
df_train_enc = df_train
df_test_enc = df_test
for col in columns_to_encode:
 encoder_dict[col].fit_transform(df_train_enc[col])

This, however, throws an error TypeError: '<' not supported between instances of 'str' and 'float'. What am I missing here? I thought LabelEncoder should be able to transform strings to numerics...

• Adonis: You might have nan values in your data, see: stackoverflow.com/q/43956705/4121573
• lte__: I don't, see updated post!

Answer 1

LabelEncoder works on string labels without an issue, so, in case you have mixed types in your data (due to missing values, for example), you can use:

for col in columns_to_encode:
 encoder_dict[col].fit_transform(df_train_enc[col].astype(str))

• Ami Tavory: Did you try astype(str)?
• lte__: Yes, that helped, will accept in 7 mins. Thank you!