March 9th, 2015 - neuro-code-review/meetings GitHub Wiki

Maxwell presented Python code for running graph community analysis on fMRI data.

Some issues/ideas that were brought up:

  • When working with NiBabel, keep in mind that get_data() will load data as int16 unless there is a scaling factor in the NIfTI header. As such, it's good to get in the habit of adding .astype('float64') (or float32 if memory usage is a concern). This will help avoid things like arithmetic overflows.

  • Docstrings should follow the NumPy/SciPy conventions as much as possible.

  • The glob module is very useful when working with unix file paths.

  • Using a linter plugin can help you catch errors as you code. Chris suggested Pyflakes, which is available as a plugin for most editors.

  • Here's a guide to "Transforming Code into Beautiful, Idiomatic Python"

  • Be mindful of changes coming in Python 3.

  • Values that do not change (constants) should be defined at the top of your code and be written in ALL_CAPS. Where possible, include a comment explaining what these values are and where they come from.

# Frequencies contributing to functional connectivity
# (Cordes et. al. 2001, ANJR)
BANDPASS = (0.01, 0.1)

  • When naming iteration variables, i should only be used when the object is an index, and even then, a name like i_scan can often be more informative.

  • In Python, is and == are not interchangeable.

  • Using continue can help avoid deeply nested loops that make code difficult to read.