Skip to content

Latest commit

 

History

History
 
 

train_and_predict

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
.flaskenv
.flaskenv
 
 
Dockerfile
Dockerfile
 
 
README.md
README.md
 
 
app.py
app.py
 
 
predict.py
predict.py
 
 
requirements-prod.in
requirements-prod.in
 
 
requirements-prod.txt
requirements-prod.txt
 
 
requirements.in
requirements.in
 
 
requirements.txt
requirements.txt
 
 
train.py
train.py
 
 

README.md

OrcaAL's ML Endpoint

OrcaAL is an active learning tool to help an orca detection model perform better.

The ML Endpoint is a small flask app that given an h5 file, a labeled dataset and an unlabeled dataset, trains an ML model on the labeled data and predicts on the unlabeled data.

Available Endpoints

Train

Trains an h5 model stored on s3 for the given number of epochs, using labeled data from s3 as well, and returns some statistics about the training process

URL: /train?model_url=[path]&labeled_url=[path]&img_width=[int]&img_height=[int]&epochs=[int]
Method: GET
Success Response:

  • Condition: If all parameters are provided, and if model_url and labeled_url paths exist on s3
  • Code: 200 OK
  • Content example:
http:https://localhost:5001/train?model_url=s3:https://orcagsoc/models/srkw_cnn_0.h5&labeled_url=s3:https://orcagsoc/labeled_test/&img_width=607&img_height=617&epochs=1

{
  "acc": [
    0.5159235596656799
  ],
  "cm": [
    9,
    11,
    10,
    23
  ],
  "labeled_files": 157,
  "loss": [
    1.031740427017212
  ],
  "model_acc": 0.6037735939025879,
  "model_loss": 0.8503243923187256,
  "model_url": "s3:https://orcagsoc/models/srkw_cnn_1.h5",
  "val_acc": [
    0.6037735939025879
  ],
  "val_loss": [
    0.8503243923187256
  ]
}

Error Response:

  • Condition: If parameters are missing, or if model_url or labeled_url paths do not exist
  • Code: 500 Server Error
  • Content example:
http:https://localhost:5001/train?model_url=s3:https://orcagsoc/models/srkw_cnn_0.h5
KeyError: 'labeled_url'

OR

http:https://localhost:5001/train?model_url=s3:https://orcagsoc/models/no_model_0.h5&labeled_url=s3:https://orcagsoc/labeled_test/&img_width=607&img_height=617&epochs=1
OSError: SavedModel file does not exist at: no_model_0.h5/{saved_model.pbtxt|saved_model.pb}

Predict

Returns the predicted value of all the unlabeled images located on an s3 bucket

URL: /predict?model_url=[path]&unlabeled_url=[path]&img_width=[int]&img_height=[int]&epochs=[int]
Method: GET
Success Response:

  • Condition: If all parameters are provided, and if model_url and unlabeled_url paths exist on s3
  • Code: 200 OK
  • Content example:
http:https://localhost:5001/predict?model_url=s3:https://orcagsoc/models/srkw_cnn_0.h5&unlabeled_url=s3:https://orcagsoc/unlabeled_test/&img_width=607&img_height=617

[
  {
    "audio_url": "https://orcagsoc.s3.amazonaws.com/unlabeled_test/mp3/orcasoundlab_1594154118.mp3",
    "duration": 3,
    "location": "Haro Strait",
    "predicted_value": 0.007200658321380615,
    "timestamp": "Tue, 07 Jul 2020 15:35:18 GMT"
  },
  {
    "audio_url": "https://orcagsoc.s3.amazonaws.com/unlabeled_test/mp3/orcasoundlab_1594154121.mp3",
    "duration": 3,
    "location": "Haro Strait",
    "predicted_value": 0.007294178009033203,
    "timestamp": "Tue, 07 Jul 2020 15:35:21 GMT"
  },
  {
    "audio_url": "https://orcagsoc.s3.amazonaws.com/unlabeled_test/mp3/orcasoundlab_1594154124.mp3",
    "duration": 3,
    "location": "Haro Strait",
    "predicted_value": 0.00517234206199646,
    "timestamp": "Tue, 07 Jul 2020 15:35:24 GMT"
  },
  {
    "audio_url": "https://orcagsoc.s3.amazonaws.com/unlabeled_test/mp3/orcasoundlab_1594154130.mp3",
    "duration": 3,
    "location": "Haro Strait",
    "predicted_value": 0.03828537464141846,
    "timestamp": "Tue, 07 Jul 2020 15:35:30 GMT"
  },
  ...
]

Error Response:

  • Condition: If parameters are missing, or if model_url or unlabeled_url paths do not exist
  • Code: 500 Server Error
  • Content example:
http:https://localhost:5001/predict?model_url=s3:https://orcagsoc/models/srkw_cnn_0.h5
KeyError: 'unlabeled_url'

OR

http:https://localhost:5001/predict?model_url=s3:https://orcagsoc/models/no_model_0.h5&unlabeled_url=s3:https://orcagsoc/unlabeled_test/&img_width=607&img_height=617
OSError: SavedModel file does not exist at: no_model_0.h5/{saved_model.pbtxt|saved_model.pb}

Getting Started

Quick Method

  • Make sure Docker is installed
  • Run:
    docker run --name activelearning_ml -d -p 5001:5001 -e AWS_ACCESS_KEY_ID=<access-key-id> -e AWS_SECRET_ACCESS_KEY=<secret-access-key> --rm jdiegors/activelearning_ml:latest

Flexible Method

  • Make sure Python and AWS CLI are installed
  • Configure AWS CLI by entering your access keys after aws configure
  • Clone the repo and cd into the project directory
  • Create a virtual environment: python -m venv venv
  • Activate the virtual environment: source venv/bin/activate or venv\Scripts\activate for windows
  • Install the dependencies: pip install -r requirements.txt
  • Run flask run to start a development server in http:https://localhost:5001

Deployment

Follow the quick start method on your server of choice.
Otherwise, to push to a different docker container registry, create an account on https://hub.docker.com, login from the command line docker login, build the image with docker build -t activelearning_ml . from within the project directory, rename it to docker tag activelearning_ml:latest <your-docker-registry-account>/activelearning_ml:latest, push it to the Docker registry docker push <your-docker-registry-account>/activelearning_ml:latest. Now you can follow the quick start method.