Skip to content

Latest commit

 

History

History
50 lines (41 loc) · 2.78 KB

logistic_regression_exercise.md

File metadata and controls

50 lines (41 loc) · 2.78 KB
orphan
true

In this exercise you will perform some simple logistic regression analysis using the new workflow (e.g. using the command line, scripts, git and github).

  • Get into teams of 2 or 3 people (smaller groups are better, no more than 3 per group)
  • One person creates a github repository called "logistic_regression_analysis"
  • Download the full CKD dataset into the repository, from here: https://github.com/matthew-brett/cfd2020/blob/master/data/ckd_full.csv
  • Everybody in the group (including the person who created the repository) creates a fork of the repository, on their own github account

Your task (as a group) is to fit the following logistic regression model Class ~ Hemoglobin. E.g. predicting whether or not a patient has CKD from their hemoglobin score.

To do this you should write the following set of scripts (as much as possible, you should each work on separate parts of the analysis, and collaborate via pull requests):

  • DATA CLEANING SCRIPT: a .py file to perform you data cleaning (e.g. creating the dummy codes for Class). Use df.to_csv() to save your clean data to a csv file with a different filename to the original file name.

  • COST FUNCTION SCRIPT: create a .py file containing the logit transformation function, the inverse logit transformation function, and the logistic regression cost function. DO NOT COPY/PASTE the functions from the textbook page, try to re-write them line-by-line yourselves so you understand what each line is doing. These functions can then can imported by other scripts in the repository, e.g. if this file is called "cost_function_utils.py" then you can import the function into other scripts using import cost_function_utils

  • DATA ANALYSIS SCRIPT: this .py file should do several things: - First, this script should plot Class as a function of Hemoglobin. - Then the script should fit your model (Class ~ Hemoglobin) using minimize (by importing your cost function from COST FUNCTION SCRIPT). - You should then fit the same model using statsmodels. - You should write some tests to check that minimize and statsmodels are producing similar parameter estimates (HINT: you may want to investigate np.isclose() and assert - ask us for help with this also). - You should then create a new plot, showing Class ~ Hemoglobin, but also showing the predictions from your logistic regression model.

If you want to, you can create a Jupyter Notebook to show the results of your data analysis script. You can use an .Rmd or .ipynb file here, as the notebook will only contain a few lines (see the example showcase_notebook.ipynb)

An example repo, with some blank templates, can be found here: https://github.com/lisds/logistic_regression_exercise