Add example notebooks + multiple major changes

[AdeelH]

Overview

This PR adds the following 5 notebooks:

• reading_raster_data.ipynb (3.89 MB)
  • Raster sources and transformers.
• reading_vector_data.ipynb (218 KB)
  • Vector sources and transformers.
• reading_labels.ipynb (1.80 MB)
  • Label sources and ClassConfig.
• sampling_training_data.ipynb (2.57 MB)
  • GeoDataset.
• misc.ipynb (2.32 KB)
  • Misc recipes. Currently only has one: "getting data from MLHub".

In service of the above, it makes the following major code changes:

• Removes ActivateMixin entirely.
• Replaces STRTree usage with GeoPandas GeoDataFrame-based spatial joins in ChipClassificaitonLabelSource and RasterizedSource.
  • This also fixes the following warnings:
    • ShapelyDeprecationWarning: STRtree will be changed in 2.0.0 and will not be compatible with versions < 2.
    • ShapelyDeprecationWarning: Setting custom attributes on geometry objects is deprecated, and will raise an AttributeError in Shapely 2.0
• Removes the mask-to-polygons dependency.
  • Because it blocked bumping up the Shapely version.
  • As part of this, also removes the "vector evaluation" in SS Evaluator which depended on it.
    • It was not generally useful enough.
• Adds numpy-like array indexing and slicing to RasterSource and LabelSource.
• Refactors x_shift and y_shift in RasterioSource(Config) into a VectorTransformer.
• Removes extent_crop added in Add ability to crop raster source extent #1030.
  • The user can now manually specify an extent instead.

Other minor changes:

• Enhances CRSTransformer so that it can now operate directly on: RV Boxes, Rasterio Windows, and Shapely geoms.
• Simplifies MultiRasterSource by removing SubRasterSourceConfig and raw_channel_order.
• Adds StatsTransformer.from_raster_sources() for easy initialization.
• Enhances RasterioSource so that it only reads the required bands.
• Disambiguates RasterSource.num_channels into RasterSource.num_channels and RasterSource.num_channels_raw.
• Adds more progress-bars
  • To geojson util functions and LocalFileSystem.
  • All progress-bars are now tqdm-based.
  • Progress-bars now have a 5 seconds delay. Which means they only show up if processing takes longer than 5 seconds. This gets rid of spurious progress-bars.
• Makes it easier to get colors in a standard format from ClassConfig.
• Refactors several getters in RasterSource and other classes into properties.
• Slightly refactors ChipClassificationLabels.
• Fixes a bug in buffer_geoms() plus other minor improvements in geojson.py.
• Misc. Learner tweaks:
  • Make scene_dataset and window_opts optional in GeoDataConfig.
  • Allow calling Learner.train() with custom number of epochs.
• Misc. Box tweaks.
• Fixes a CI bug by pinning pyopenssl version.
  • See https://askubuntu.com/questions/1428181/module-lib-has-no-attribute-x509-v-flag-cb-issuer-check.
• Remove no-longer-needed patch for torch.hub (External Model Definition Functionality Potentially Broken in PyTorch 1.9 #1271).
• Updates the grep regex in Dockerfile to handle blank lines and comment lines.

Checklist

• [ ] Added needs-backport label if PR is bug fix that applies to previous minor release
• Ran scripts/format_code and committed any changes
• Documentation updated if needed
• PR has a name that won't get you publicly shamed for vagueness

Notes

Requirements changes:

• mask-to-polygons removed
• Shapely bumped to ==1.8.4.
• GeoPandas and PyGEOS added.

Backward-compatibility breaking changes:

• Removal of SubRasterSourceConfig.
• Refactoring of x/y_shift.
• Make background_class_id required in ChipClassificationLabelSourceConfig if inferring cells.

Testing Instructions

• See updated unit tests.
• See notebooks.

Closes #1275
Part of #1460
Part of #1459

[codecov]

Codecov Report

Merging #1470 (4b022ca) into master (5a1e6f1) will decrease coverage by 1.65%.
The diff coverage is 66.02%.

@@            Coverage Diff             @@
##           master    #1470      +/-   ##
==========================================
- Coverage   73.35%   71.70%   -1.66%     
==========================================
  Files         184      186       +2     
  Lines        8884     8947      +63     
==========================================
- Hits         6517     6415     -102     
- Misses       2367     2532     +165     

Impacted Files	Coverage Δ
...stervision_core/rastervision/core/data/__init__.py	100.00% <ø> (ø)
...aluation/semantic_segmentation_evaluator_config.py	40.90% <0.00%> (-7.37%)	⬇️
rastervision_core/rastervision/core/predictor.py	19.40% <ø> (+0.28%)	⬆️
...stervision/core/rv_pipeline/chip_classification.py	40.74% <0.00%> (ø)
.../rastervision/core/rv_pipeline/object_detection.py	21.11% <0.00%> (ø)
...ervision/core/rv_pipeline/semantic_segmentation.py	18.27% <0.00%> (ø)
.../pytorch_backend/pytorch_learner_backend_config.py	66.00% <0.00%> (+1.29%)	⬆️
.../pytorch_learner/dataset/classification_dataset.py	92.59% <ø> (ø)
...ytorch_learner/dataset/object_detection_dataset.py	33.65% <ø> (ø)
...a/label_store/semantic_segmentation_label_store.py	25.12% <7.40%> (-1.22%)	⬇️
... and 66 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[lewfish]

Your clarifications made sense. I think we should leave things as is.