Go through the task_descriptor.txt file to understand the task and the dataset. You can summarize it in your research logs to keep track of what all you have to do.
Then fill in the provided train.py script to train a model and iterate over different models or feature selections to get a better performance.
Never try to read any csv files directly. Do not forget to execute the changes you made to check for performance.