It is worth your effort to standardize on repository structures to make analysis/collaboration/reproducibility easier.
- See https://drivendata.github.io/cookiecutter-data-science/ for a datascience oriented structure. It doesn't always apply for every economics project, but follow it if you can
- A bigger framework for setting up solid ML/datascience pipelines is https://kedro.readthedocs.io/en/stable/
- Do not invent your own structure!
For displaying graphs/output/etc. inline, create a file
import matplotlib.pyplot as plt
import numpy as np
x = np.linspace(0, 20, 100) # Create a list of evenly-spaced numbers over the range
plt.plot(x, np.sin(x)) # Plot the sine of each x point
plt.show() And then <Ctrl-Shift-P> to enter the command, and choose > Run Current File in Python Python Interactive window to see plots in the interactive terminal.
Alternatively, you can run the file directly in normal terminal
- You may see the error
qt.qpa.plugin: Could not load the Qt platform plugin "windows" in "" even though it was found. - If so, then there may be changes required. Possible miktex collision setting the
QT_PLUGIN_PATH? See https://stackoverflow.com/questions/41994485/error-could-not-find-or-load-the-qt-platform-plugin-windows-while-using-matplo
- When
code blackis installed as the default in vscode,<Ctrl-Shift-P>to findFormator chooseShift-Alt-Fto format with the standard <Shift-Enter>in a.pyfile will run the python in an interactive window or terminal
- Pytorch discourse: https://discuss.pytorch.org/
- There is also a slack for more advanced questions
- Unit testing package: https://github.com/suriyadeepan/torchtest
- VSCode Extension: https://marketplace.visualstudio.com/items?itemName=SBSnippets.pytorch-snippets
- Torchscript (i.e. auto-differentiable python subset
- Tutorials