Documentation bugs

This is a running list of bugs I’ve noticed in the documentation that need to be fixed.

• Sections don’t have permalinks
• “Open in Colab” button is too wide: https://github.com/pytorch/pytorch_sphinx_theme/issues/140, https://github.com/pytorch/pytorch_sphinx_theme/pull/142
• Fonts are missing, requiring hack to requirements.txt: https://github.com/pytorch/pytorch_sphinx_theme/issues/143
• Dataset base class is incorrect: https://github.com/sphinx-doc/sphinx/issues/9607
• Models and trainers are missing args/returns documentation
• Tutorials don’t run on Colab, need to clone from GitHub and install dependencies (or pip install once we release on PyPI)
• Notes on how to download SEN12MS and So2Sat aren’t showing up (https://github.com/microsoft/torchgeo/commit/3d3b903d94c9ada2f9e0e919f254d7a8406d7e25, https://github.com/microsoft/torchgeo/commit/fd4998b78737aad6924a7d67c912f01661f96839)
• API docs show “Return type: None” for most functions, see if we can get rid of this (#277)
• In-line code missing off-color background (https://github.com/pytorch/pytorch_sphinx_theme/issues/130)
• the “well-tested” and “linters” links are broken on the users/contributing page (6d5725056c2c7851e2dc9a47952e2cd884ed43a4)
• NAIP has no docs because docstrings aren’t inherited from parent classes
• BoundingBox docs aren’t formatted correctly: https://torchgeo.readthedocs.io/en/latest/api/datasets.html#torchgeo.datasets.BoundingBox
• Nested lists have a lot of unnecessary whitespace
• Models have extra undocumented parameters defined
• Trainer kwargs are showing up under parameters
• Versions popup menu appears on wrong side of the screen
• Versions popup menu does not have release versions, only latest and stable
• __getitem__ is showing up before __init__ because things are sorted alphabetically, we don’t want this
• torchgeo.samplers.Units: possible enum values not displayed
• torchgeo.datasets.RasterDataset: class attributes not displayed

[calebrob6]

Thanks for the catch / report @robmarkcole! This is one of the few things that isn’t tested… Fixed in #434

[robmarkcole]

There is an error in trainers.ipynb. The csv is created with upper case column names (e.g. train_RMSE). Cell fixed below:

train_steps = []
train_rmse = []

val_steps = []
val_rmse = []
with open(os.path.join(experiment_dir, "tutorial_logs", "version_0", "metrics.csv"), "r") as f:
    csv_reader = csv.DictReader(f, delimiter=',')
    for i, row in enumerate(csv_reader):
        try:
            # train_rmse.append(float(row["train_rmse"]))
            train_rmse.append(float(row["train_RMSE"]))
            train_steps.append(i)
        except ValueError: # Ignore rows where train RMSE is empty 
            pass

        try:
            # val_rmse.append(float(row["val_rmse"]))
            val_rmse.append(float(row["val_RMSE"]))
            val_steps.append(i)
        except ValueError: # Ignore rows where val RMSE is empty
            pass