Implementation Notebooks

[ebgoldstein]

put some example notebooks online

[ebgoldstein]

• [x] single prediction
• [x] prediction confidence
• [x] TTA
• [ ] batch prediction

[dbuscombe-usgs]

• [ ] Otsu threshold versus simple threshold (binary problems)
• [ ] speed up inference using concurrent futures e.g. https://stackoverflow.com/questions/66036272/predicting-in-parallel-using-concurrent-futures-of-tensorflow-keras-models

[dbuscombe-usgs]

• [ ] simple ensemble models (multiple models trained for the same task)
• [ ] more complex ensemble models (models trained for different tasks, e.g. a binary model, and a multiclass model containing the same class as the binary one)

[dbuscombe-usgs]

• [ ] download a Zenodo model (model selected using a widget) and use it for prediction on a Landsat image

[dbuscombe-usgs]

This needs a small overhaul. I think it would be good to use each notebook to demonstrate a basic usage. Each notebook would use a different zoo model, for illustration. Ideally, the notebooks should be reproducible, which is made possible by adding the functionality for selecting/downloading a model using zenodo

I propose the following changes

1. @dbuscombe-usgs to update EnsemblePrediction_2class by selecting/downloading updated zoo model
2. @dbuscombe-usgs to update and rename HyperEnsemblePrediction_binary_and_multiclass_model_merging by selecting/downloading updated zoo model (replace Elwha example with another model)
3. @ebgoldstein to update SingleImagePrediction by selecting/downloading NOAA zenodo model (when updated). This would use the 'BEST_MODEL.txt' convention

I just realized that you can't make "Zenodo" without "Zoo", but really it is the other way around

[dbuscombe-usgs]

I'm closing this. My focus is on making tools (scripts), not examples. Feel free to reopen if/when we need