Running your first experiment

Quickstart

1. Instrumenting a python script

from snapper_ml import job

import numpy as np
from sklearn.svm import SVC
from sklearn.model_selection import train_test_split
from sklearn.datasets import load_iris

SEED = 1234


@job  # ADD THIS DECORATOR
def main(C=1.0, kernel='rbf', degree=3, gamma='scale'):
    np.random.seed(SEED)
    iris = load_iris()
    X_train, X_val, y_train, y_val = train_test_split(iris.data, iris.target, random_state=SEED)
    model = SVC(C=C, gamma=gamma, kernel=kernel, degree=degree)
    model.fit(X_train, y_train)
    accuracy = model.score(X_val, y_val)
    return {'val_accuracy': accuracy}


if __name__ == '__main__':
    main()

2. Defining a configuration file

Defining a configuration file is straightforward, we just need to create a YAML/JSON file and specify the name of the experiment, its parameters, and the file to execute.

name: "SVM #1"

params:
  C: 10
  kernel: poly
  gamma: auto
  degree: 3

run:
  - examples/scripts/train_svm.py

3. Executing the experiment

Finally, once we have an instrumented Python script and a config file, we can execute the job as follows:

snapper-ml --config_file examples/experiments/svm.yaml

Running jobs in a Docker container

docker_config:
  image: image_name:tag

docker_config:
  dockerfile: path/to/dockerfile

Adding callbacks

Callbacks allow one to hook into the process and react to events. snapper-ml offers a simple callback system based on the class Callback. In order to implement a custom callback, all it takes is implementing that abstract class.

Once we’ve implemented a custom callback class, we can add an instance of it as callbacks argument of job. The module notifiers contains some predefined callbacks for the most common scenarios.

from snapper_ml import job
from snapper_ml.callbacks.notifiers import DesktopNotifier

import numpy as np
from sklearn.svm import SVC
from sklearn.model_selection import train_test_split
from sklearn.datasets import load_iris

SEED = 1234


# ADD THIS
@job(callbacks=[DesktopNotifier()])
def main(C=1.0, kernel='rbf', degree=3, gamma='scale'):
    np.random.seed(SEED)
    iris = load_iris()
    X_train, X_val, y_train, y_val = train_test_split(iris.data, iris.target, random_state=SEED)
    model = SVC(C=C, gamma=gamma, kernel=kernel, degree=degree)
    model.fit(X_train, y_train)
    accuracy = model.score(X_val, y_val)
    return {'val_accuracy': accuracy}


if __name__ == '__main__':
    main()