chris-lovejoy / codingformedicine Goto Github PK

@chris-lovejoy and I have had some issues with getting notebooks to run due to differing pandas versions etc

Concretely, here are my relevant packages:

(coding_for_medicine)  [truncated] > conda list | grep -E 'pandas|numpy|seaborn|matplotlib|imbalanced-learn'
imbalanced-learn          0.9.1                    pypi_0    pypi
matplotlib                3.6.1           py310h2ec42d9_0    conda-forge
matplotlib-base           3.6.1           py310he725631_0    conda-forge
matplotlib-inline         0.1.6              pyhd8ed1ab_0    conda-forge
numpy                     1.23.4          py310h1b7c290_0    conda-forge
pandas                    1.5.1           py310hecf8f37_0    conda-forge
seaborn                   0.11.2               hd8ed1ab_0    conda-forge
seaborn-base              0.11.2             pyhd8ed1ab_0    conda-forge

Should we pin these to a specific version, and have a prompt to install the right version on each run?

Also are we using conda or pip? If using Colab I imagine this doesn't really matter ~ but good to get uniformity across notebooks

@chris-lovejoy Do you have a particular idea of how you want students to get data?

My current method uses the kaggle CLI, however I thought it was surprised by how much of a faff it is to use (having to create an account, manually pasting a .json file into ~/.kaggle etc.) - not beginner friendly in my opinion.

Did you invision just including the data within the repo, or having the notebook grab it from a source?

Not hugely important but I suppose it would be ideal to have a uniform approach between notebooks.

The breast cancer features exercise is only 'loosely' interactive. It encourages users to manipulate the code, but has no actual requirements for this and can run front-to-back without modification. Therefore, this exercise does not currently pass Principle 2 (of being interactive).

To improve it, there should be more explicit "stop points" where users are prompted to modify the code. For example, this could describe functions that could be used, but which aren't yet implemented - and invite the user to implement them and play around.

I will look into making the 'easy' python stuff :)

after training the model using Randomforest classifier i have added a simple GUI using tkinter in python, the GUI takes random user inputs based on 3 predection values and delivers then an Output M or B for the daignosis.

remove number-based sorting of exercises

create a table that lists all exercises, what technologies they include, how difficult, links to the respective notebook, etc

(also add an intro on the README detailing how to get started with any of the notebooks)

The simplest way to do this will be create a pip virtual env (as per current readme descriptions), install the packages, and make sure that all notebooks run without problem -> if so, then can save as requirements.yml file and add a description to the readme for installing those dependencies

Start work on Stroke Prediction exercise at over here (file)

Rough outline for things to cover/notebook structure:

• Get data / load into runtime / imports etc.
• Data Exploration
• Build models
• Explore Accuracy Paradox
  • Look at better alternatives, e.g. F1/AUC/etc
• Overfitting and underfitting

Anything else worth going into @chris-lovejoy ?

Is it possible to be added as a collaborator to the repo?