Python (and R) with KNIME can be useful if there are functions like LogLoss that are not (yet) present within KNIME and sometimes a task can be easier in Python than with a lot of KNIME nodes.
But as has been mentioned: you do not have to choose you can use both 
A few examples of use of Python:
- you will need Python if you want to use the KNIME Keras integration
- you can further manipulation existing Excel files with Python
- a complex group of group task that also could be done in KNIME
- use SQLite to transfer data to Python
- and you can use all the predictive models available in Python. You will have to see how a lot of them handle categorical data