Document Segmentation Model Evaluation (#544)

CHANGELOG.md

@@ -13,6 +13,7 @@ created.
 ## Upcoming
 
 ### Added
+- ([#544](https://github.com/microsoft/InnerEye-DeepLearning/pull/544)) Add documentation for segmentation model evaluation.
 - ([#465](https://github.com/microsoft/InnerEye-DeepLearning/pull/465/)) Adding ability to run segmentation inference
 module on test data with partial ground truth files. (Also [522](https://github.com/microsoft/InnerEye-DeepLearning/pull/522).)
 - ([#502](https://github.com/microsoft/InnerEye-DeepLearning/pull/502)) More flags for fine control of when to run inference.

docs/move_model.md

@@ -10,14 +10,19 @@ training time:
 
 If you want to export a model from one Workspace to another you can use the following command to download and upload a
 model from an AzureML workspace. This script does not use settings.yml, it uses interactive authentication, and the
-workspace specified in the parameters. The model will be written to the path in --path parameter with two folders one
-for the `MODEL` and one for the `ENVIRONMENT` files.
+workspace specified in the parameters.
 
-- Download to
-  path: `python InnerEye/Scripts/move_model.py -a download --path ./ --workspace_name "<name>" --resource_group "<name>" --subscription_id "<sub_id>" --model_id "name:version"`
+- Download to path:
 
-- Upload from
-  path: `python InnerEye/Scripts/move_model.py - upload --path ./ --workspace_name "<name>" --resource_group "<name>" --subscription_id "<sub_id>" --model_id "name:version"`
+`python InnerEye/Scripts/move_model.py -a download --path ./ --workspace_name "<name>" --resource_group "<name>" --subscription_id "<sub_id>" --model_id "name:version"`
+
+ The model will be downloaded to a subfolder called `name_version` (based on the `--model_id` parameter) of the path in `--path` parameter. For example, with `--model_id "my_model:100` this will be `my_model_100`. The subfolder will contain two folders: one for the `MODEL` and one for the `ENVIRONMENT` files.
+
+- Upload from path:
+
+`python InnerEye/Scripts/move_model.py -a upload --path ./ --workspace_name "<name>" --resource_group "<name>" --subscription_id "<sub_id>" --model_id "name:version"`
+
+The model is expected to be in the format above, i.e. contained in a subfolder called `name_version` of the path in `--path` parameter. The model will be automatically given the next available `id` once uploaded to the workspace. For example, if `my_model:100` already exists then it will be called `my_model:101`.
 
 Once in place you may want to run inference on the model with partial test data, i.e. test data from patients for whom
-some of the labels are missing. Normally inference on partial test data would raise an exception. To allow inference to continue over partial test data add the flag `--allow_incomplete_labels` to your inference call, for example `python InnerEye/ML/runner.py --allow_incomplete_labels --train=False --azureml --model=<model_name> --run_recovery_id=<recovery_id>`
+some of the labels are missing. Normally inference on partial test data would raise an exception. To allow inference to continue over partial test data add the flag `--allow_incomplete_labels` to your inference call, for example `python InnerEye/ML/runner.py --allow_incomplete_labels --no-train --azureml --model=<model_name> --run_recovery_id=<recovery_id>`