mltrics 0.1.4

Monitor ML models in production

Mltrics

Official command line utility to use Mltrics API programmatically.

We help businesses evaluate, compare, and monitor machine learning models in production. Therefore, identify failure cases and take action immediately.

Installation

Mltrics and its required dependencies can be installed using pip:

pip install mltrics

Usage

Once mltrics package is installed, check out the following usage documentation:

Authenticate to mltrics platform

from mltrics.mltrics import MltricsClient

import getpass

username = input("Enter username: ")
password = getpass.getpass(prompt='Enter password: ')
client = MltricsClient(username=username, password=password, env="prod")

Update user profile

organization, full_name = "<your-organization-name>", "<Your name>"
client.update_user_profile(organization=organization, full_name=full_name)

Create model and upload predictions

Create a baseline model

baseline_model_id, baseline_model_name = 'nn_iter_10k', 'Neural network (trained for 10K iters)'
baseline_model = client.create_model(baseline_model_id, baseline_model_name)
print(baseline_model)

Update model details

baseline_model_new_name = "Neural network (trained for 20K iters)"
updated_model = client.update_model_details(baseline_model_id, baseline_model_new_name)
print(updated_model)

Get uploaded models

models = client.get_models()
models

Upload predictions for model

baseline_preds = [
     {
      'pred_class': 'dog',
      'label_class': None,
      'model_id': baseline_model_id,
      'image_id': 'img1',
      'image_url': 'https://mltrics.s3.us-west-2.amazonaws.com/datasets/cats_vs_dogs/Cat/10896.jpg',
      'pred_file': None,
      'predictions': {},
     },
]
response = client.upload_model_predictions(baseline_model_id, baseline_preds)
print(response)

See all uploaded predictions

predictions = client.get_model_predictions(baseline_model_id)
predictions

Create candidate model and upload predictions for candidate model

### Create candidate model and upload predictions

candidate_model_id, candidate_model_name = "nn_50k_iter", "Neural Network (50K iter)"
candidate_model = client.create_model(candidate_model_id, candidate_model_name)

candidate_preds = [
     {
      'pred_class': 'cat',
      'label_class': None,
      'model_id': candidate_model_id,
      'image_id': 'img1',
      'image_url': 'https://mltrics.s3.us-west-2.amazonaws.com/datasets/cats_vs_dogs/Cat/10896.jpg',
      'pred_file': None,
      'predictions': {},
     },
]
response = client.upload_model_predictions(candidate_model_id, candidate_preds)
print(response)

Get model comparison between baseline and candidate model

comparison_results = client.compare_model_predictions(baseline_model_id, candidate_model_id)
print(comparison_results)