Description

Uniovi Simur WearablePerMed Predictor.

Build and publish docker image

You must install docker previous to use these commands

To build the image execute this command:

$ docker build -t wearablepermed-predictor:1.18.0 .

To tag the image to be published in simuruo docker hub account execute this command:

$ docker build tag wearablepermed-predictor:1.18.0 simuruo/wearablepermed-predictor:1.18.0 

Login in simuruo docker hub account execute this command:

$ docker login -u simuruo
Password: 

To publish image in simuruo docker hub account execute this command:

$ docker push simuruo/wearablepermed-predictor:1.18.0

Execute from Python package

If you want use our models published in Huggin Face you must execute a command like this:

$ predictor \
--model-type randomforest \
--sensor-channel accelerometer_gyroscope \
--segment-body wrist \
--class-type classes_4 \
--resources-folder /home/miguel/temp/models/wearablepermed_models/input \
--resource-id PMP1024_W1_M.csv \
--cases-folder /home/miguel/temp/models/wearablepermed_models/results \
--case-id case_M_BRF_acc_gyr_4_classes \
--case-file-format csv \
--verbose

If you want to use your custom models must execute a command like this:

$ predictor \
--model_file /home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/models/RandomForest.pkl,
--label_file /home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/models/label_encoder.pkl,
--resources-folder /home/miguel/temp/models/wearablepermed_models/input,
--resource-id PMP1024_W1_M.csv,
--cases-folder /home/miguel/temp/models/wearablepermed_models/results,
--case-id case_M_BRF_acc_gyr_4_classes,
--case-file-format csv,
--verbose

Actually only Ranfom Forest is implemented in predictor

Execute from docker image

$ docker run \
--rm \
-u $(id -u):$(id -g) \
-v /home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/data/input:/home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/data/input \
-v /home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/data/output:/home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/data/output \
simuruo/wearablepermed-predictor:1.18.0 \
--model-type randomforest \
--sensor-channel accelerometer_gyroscope \
--segment-body wrist \
--class-type classes_4 \
--resources-folder /home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/data/input \
--resource-id case_PI_BRF_acc_gyr_01/PMP1024_W1_PI_1.csv \
--cases-folder /home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/data/output \
--case-id case_PI_BRF_acc_gyr_4_classes_01 \
--verbose

Predictor arguments

These arguments are used if you select Python Package or Docker containers to execute predictor command:

• model-type: The model type: esann, capture24, randomforest, xgboost.

• sensor-channe: The model channels trained: accelerometer, accelerometer_gyroscope.

• class-type: The model class type trained: classes_4, classes_15.

• segment-body (*): The model segment body trained: thigh, wrist, hip.

• model-file: if you want use your custom model set the model file

• label-file: if you want use your custom labels set the label file

• resources-folder (*): The root resourcers folder.

• resource-id (*): The resource file id in csv format.

• cases-folder: The root cases folder.

• case-id: Case unique name where save results under cases-folder.

• case-file-format: Case file format. Default is npz. Possible values: [npz, csv].

• is-label-export: Specify if predictions are export as label format. Default is False.

• is-database-export: The prediction result is database saved. Default is False.

• verbose: activate verbose logging mode.

(*) are mandatory arguments

If you want login inside the container execute this command.

$ docker run \
--rm \
-it \
-u $(id -u):$(id -g) \
-v /home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/data/input:/home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/data/input \
-v /home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/data/output:/home/miguel/git/uniovi/simur/uniovi-simur-wearablepermed-predictor/data/output \
--entrypoint sh \
simuruo/wearablepermed-predictor:1.18.0

Default Value

All models offered by predictor are trained with

• Window size of 250 and overlapping of 50%.
• Right now only individual models are offered by predc¡ictor: Wrist, Thigh or Hip segment bodies.

Build and Publish in Pypi and Docker Hub

1. Set the final version to the precitor python package from file setup.cfg

   version = 1.18.0
   

2. Set the new version in the shell scripts: run_predictor.sh, run_predictor.bat

   Linux/Mac run_predictor.sh script:

   # --- CONFIGURATION (Change these) ---
   PREDICTOR_VERSION="1.18.0"
   

   Windows `run_predictor.bat script:

   :: --- SYSTEM CONFIGURATION ---
   set PREDICTOR_VERSION=1.18.0
   

3. Rebuild and publish package in Pypi repository (You must have credentials)

   $ tox -e clean
   $ tox -e build
   $ tox -e publish -- --repository pypi
   

4. Finally build docker image with the last version selected and publish in simuruo Docker Hub account (You must have credentials)

   $ docker build -t wearablepermed-predictor:1.18.0 .
   $ docker tag wearablepermed-predictor:1.18.0 simuruo/wearablepermed-predictor:1.18.0
   $ docker push simuruo/wearablepermed-predictor:1.18.0
   

Note

This project has been set up using PyScaffold 4.6. For details and usage information on PyScaffold see https://pyscaffold.org/.