roofAI 🏛️ meets ⚒️ SageMaker + more-29

Next, a large train,

roofAI conda activate sagemaker

roof conda activate sagemaker

@select roofAI-sagemaker-dataset-AIRS-1000-v1

roofAI dataset ingest \
    source=AIRS,target=sagemaker,upload . \
    --test_count 0 \
    --train_count 1000 \
    --val_count 200

@select roofAI-sagemaker-semseg-AIRS-1000-v1

sagesemseg train - .. . \
    --instance_type ml.g4dn.2xlarge

@browse sagemaker dashboard

accuracy of 1.0 and mIOU=0.5 may indicate conversion issues.

as expected. 🛑 count issue?

roof conda activate sagemaker

@select roofAI-sagemaker-dataset-AIRS-1000-v4

roofAI dataset ingest \
    source=AIRS,target=sagemaker,upload . \
    --test_count 0 \
    --train_count 1000 \
    --val_count 200

@select roofAI-sagemaker-semseg-AIRS-1000-v4

sagesemseg train - .. . \
    --instance_type ml.g4dn.2xlarge

numbers make sense.

@sagemaker browse dashboard

@select roofAI-sagemaker-semseg-AIRS-1000-v4
@download filename=validation/output.jpg .

local test,

roofAI dataset ingest \
  source=AIRS,target=sagemaker - \
  --test_count 16 \
  --train_count 16 \
  --val_count 16

obsolete.

publish giza on pypi, validate including .sh files.

create main and leave master for the rpis?

Next: ci/cd for 🔻giza.