Minimal Conda environment for locally running the Coursera ML Specialization labs

coursera-ml.environment.yml

#
# Minimal Conda environment for locally running the Coursera Machine Learning Specialization
# labs from https://www.coursera.org/specializations/machine-learning-introduction
#
# To use, download and unpack the notebooks from the course, make sure you have conda
# installed (Miniconda is fine), then create the environment for the first time:
#
# > conda env create -f coursera-ml.environment.yml
#
# Some optional labs, like Decision Trees in Course 2, Week 4, require graphviz
# to be installed for certain visualizations. If you want to run these, you must install
# graphviz separately, or you'll get an error about "dot" not being in the PATH at runtime.
# For example, in Ubuntu, run:
#
# > sudo apt install graphviz
#
# After that, anytime you want to launch the labs, cd to the directory you've downloaded
# them to, run the following, and launch the url it displays.
#
# > conda activate coursera-ml
# > jupyter lab
#
---
name: coursera-ml

channels:
  - defaults
  - conda-forge

dependencies:

  # These versions match exactly what Coursera's Jupyter environment was running
  # for the Machine Learning Specialization labs as of 2023/12/28.
  - gcc=13.2.0
  - matplotlib=3.3.2
  - networkx=2.6.3
  - numpy=1.21.6
  - pandas=1.0.3
  - python=3.7.6
  - scikit-learn=0.22.2.post1
  - sympy==1.5.1
  - tensorflow=2.8.0

  # It's not critical to match the exact version of JupyterLab Coursera uses,
  # and we prefer something stable and up-to-date. This is the latest 3.x version.
  - jupyterlab=3.6.6

  # This provides support for %matplotlib magic in notebooks. Coursera runs
  # ipympl 0.5.6, but this is only compatible with Jupyterlab 2.x according to
  # https://matplotlib.org/ipympl/installing.html#compatibility-table
  # Since we want JupyterLab 3.x, here we diverge and use the latest version
  # compatible with that and Matplotlib 3.3.2.
  - ipympl=0.8.8

  # Coursera uses 7.5.1, but that's not compatible with ipympl 0.8.8,
  # which needs at least 7.6. So here we choose the latest 7.6.x available.
  - ipywidgets=7.6.6

  - pip
  - pip:
    # We install this via pip because this way it doesn't try to install
    # graphviz via conda, which would fail because of an incompatible version
    # of libffi it requires (the version of Python we're using requires an older
    # version of that library). Doing it this way, we rely on "dot" being in
    # the PATH at runtime, as described above.
    - pydot==1.4.2

variables:

  # Not required, but this keeps TensorFlow warnings/info messages to a minimum
  TF_CPP_MIN_LOG_LEVEL: 2