Convert column text data into features using python to use for machine learning

Question

Left-side CSV file has five columns .application column has several app types delimited with ;. Depending on the app, device and district type, I want to predict the target. But I want to first convert the file in to the right-hand side dataframe to apply machine learning.

How can I do this using python?

Answer 1

You need to apply multi-hot encoding for application column and one hot encoding for other columns.

Here is my solution!

>>> import pandas as pd
>>> import numpy as np

>>> df = pd.DataFrame({'number': np.random.randint(0,10,size=5),
                  'device': np.random.choice(['a','b'],size=5),
                  'application': ['app2;app3','app1','app2;app4', 'app1;app2', 'app1'],
                  'district': np.random.choice(['aa', 'bb', 'cc'],size=5)})

>>> df

    application device  district    number
0   app2;app3   b         aa    3
1   app1        a         cc    7
2   app2;app4   a         aa    3
3   app1;app2   b         bb    9
4   app1        a         cc    4

from sklearn.preprocessing import OneHotEncoder, MultiLabelBinarizer

mlb = MultiLabelBinarizer()
# Assuming appl names are separated by ;
mhv = mlb.fit_transform(df['application'].apply(lambda x: set(x.split(';'))))
df_out = pd.DataFrame(mhv,columns=mlb.classes_)

enc = OneHotEncoder(sparse=False)
ohe_vars = ['device','district'] # specify the list of columns here
ohv = enc.fit_transform(df.loc[:,ohe_vars])
ohe_col_names = ['%s_%s'%(var,cat) for var,cats in zip(ohe_vars, enc.categories_) for cat in cats]

df_out.assign(**dict(zip(ohe_col_names,ohv.T)))

df_out