Demo-1 Sentinel-2 Random Forest Classifier - Supervised classification

This demo is based on the exercise WEkEO ai4EM MOOC - Supervised classification using Sentinel-2 data

This demo is a starting point for the project activities and introduces the potential of the Common Worflow Language for training and infer activities and MLflow.

Use the Visual Studio development container to run this demo.

Training

Using MLFlow:

mlflow run --env-manager local \
    -P training_water=./training_data/water.txt \
    -P training_artificial=./training_data/artificial_surfaces.txt \
    -P training_low_vegetation=./training_data/low_vegetation.txt \
    -P training_tree_cover=./training_data/tree_cover.txt \
    -P validation=./validation_data/validation_points.txt \
    -P img_folder=./S2_data
    .

Using a Common Workflow Language runner:

cwltool --no-container \
    train.cwl \
    --environment environment.yml \
    --train train.py \
    --ml_project MLproject \
    --max_depth 10 \
    --n_estimators 5 \
    --random_state 0 \
    --s2_data ./S2_data \
    --train_artificial_surfaces ./training_data/artificial_surfaces.txt \
    --train_low_vegetation ./training_data/low_vegetation.txt \
    --train_tree_cover ./training_data/tree_cover.txt \
    --train_water ./training_data/water.txt \
    --validation ./validation_data/validation_points.txt

Inference

The inference takes the Sentinel-2 acquistion to classify with the provided model.

CWL can be used to do so with:

cwltool --no-container \
    infer.cwl \
    --infer infer.py \
    --model_directory mlruns/0/ \
    --s2_data $PWD/S2_data/ \
    --model_id 593f7ddb798e49d0818e394d0b214b70 

Build the inference docker container with the selected model id:

docker build --build-arg model_id=593f7ddb798e49d0818e394d0b214b70 -f Dockerfile.infer .