Optional tools

Widely used general ML tools

R, caret - general ML library, similar to scikit-learn. Supports also Keras as back-end for deep learning. Supports parallel computing.
PyTorch, deep learning framework
Dask-ML, scalable machine learning with Scikit-Learn, XGBoost, and others.
- Dask+ scikit-learn example

CAST - improve spatial-temporal modelling tasks using 'caret'.
- OpenGeoHub, Spatial machine learning for GIS with R course materials

Options:
- ArcGIS Pro, very easy to use with default settings / existing models
- ArcGIS Python API, ArcGIS Notebooks, model traiing, more advanced options.
- ArcGIS Image server, Trained models at scale in production
Shallow learning: K-means, SVM, random forest, maximum likelihood classifications and ISO clustering
Deep learning:
- object detection: find bbox of the objects
- pixel classification
- object classification: classify features or tiles
- Based on PyTorch and Keras
Export Training Data For Deep Learning, 5 different formats, “Classified tiles” similar to our exercise. Very easy to use, but clearly slower than GDAL
Training serious models requires GPU, either power-PC or GPU server, not suitable for CSC GPU resources:
- ArcGIS Pro only as Windows software
- ArcGIS Python API could be installed, but no access to local data

Plugins:

Semi-Automatic Classification Plugin, allows for the supervised classification of remote sensing images, providing tools for the download, the preprocessing and postprocessing of images. https://fromgistors.blogspot.com/
A few new / little used ones, also for data preparation: machine-learning, deep-learning

pointnet (++) → Deep Learning for pointclouds directly:https://github.com/charlesq34/pointnet
RAPIDS - GPU accelerated data science, including cuspatial (point-in-polygon, spatial join, coordinate systems, shape primitives, distances, and trajectory analysis) -> check out their example notebook
ERDAS imagine
Google Earth Engine
GRASS, r.learn.ml, supervised classification and regression of GRASS rasters using the python scikit-learn package
McFly, build and compare different DL models for timeseries classification task
PyspatialML, for applying scikit-learn machine learning models to 'stacks' of raster datasets.
Orfeo Toolbox. Looks good from documentation, but in practice had several problems...
OTBTF: Orfeo ToolBox meets TensorFlow
Torchgeo
Geotorch
More links: https://github.com/sacridini/Awesome-Geospatial#deep-learning
Point cloud voxelizer example: https://github.com/Eayrey/3D-Convolutional-Neural-Networks-with-LiDAR/blob/master/point_cloud_voxelizer.py

Puhti documentation: