Example application which leads all over training process via Python CLI

app.py

import logging
import os
import sys
import time

from auger.api.dataset import DataSet
from auger.api.model import Model
from auger.api.project import Project
from auger.api.experiment import Experiment, AugerExperimentSessionApi
from auger.api.utils.context import Context
from auger.cli.utils.config import AugerConfig


PREDICTION_SOURCE = 'files/iris_data_test.csv'
PREDICTION_TARGET = 'files/irist_data_set_predict.csv'
DATASET_NAME = 'iris.csv'

log = logging.getLogger('auger.ai')


class ExampleApp():
    def __init__(self, ctx):
        self.ctx = ctx
        def print_(*args, **kwargs):
            print(*args, **kwargs)
        self.ctx.log = print_
        self.dataset = None
        self.model_id = None
        self.project = Project(
            self.ctx, self.ctx.get_config('auger').get('project', None))
        self.project.create()

    def _get_datasets(self):
        """return list of existing datasets in the Aguer Cloud"""
        dataset_list = []
        for dataset in iter(DataSet(self.ctx, self.project).list()):
            dataset_list.append(dataset['name'])
        return dataset_list

    def _start_experiment(self, experiment_name):
        experiment = Experiment(self.ctx, self.dataset, experiment_name)
        experiment_name, session_id = \
            experiment.start()
        AugerConfig(self.ctx).set_experiment(experiment_name, session_id)
        return experiment, session_id

    def prepare_dataset(self):
        """check whether dataset selected, if not, select or create one"""
        self.ctx.log("Checking dataset...")
        selected = self.ctx.get_config('config').get('dataset', None)
        if selected is None:
            if DATASET_NAME in self._get_datasets():
                # try to select existing
                AugerConfig(self.ctx).set_data_set(
                    DATASET_NAME, '').set_experiment(None)
                self.dataset = DataSet(self.ctx, self.project, DATASET_NAME)
            else:
                # or create new
                self.ctx.log("No dataset found, creating the first one...")
                source = self.ctx.get_config('config').get('source', None)
                dataset = DataSet(self.ctx, self.project).create(source)
                AugerConfig(self.ctx).set_data_set(
                    dataset.name, source).set_experiment(None)
                self.dataset = DataSet(self.ctx, self.project, dataset.name)
        else:
            self.dataset = DataSet(self.ctx, self.project, selected)
            self.ctx.log("Currently selected: %s" % selected)

    def run_experiment(self):
        experiment_name = self.ctx.get_config('auger').get(
            'experiment/name', None)
    
        self.experiment, run_id = self._start_experiment(experiment_name)

        self.ctx.log("waiting for experiment %s to finish" % experiment_name)

        self.experiment.wait(run_id)
        leaderboard, status = self.experiment.leaderboard(run_id)
        self.model_id = leaderboard[0]['model id']

    def deploy(self):
        Model(self.ctx, self.project).deploy(self.model_id, locally=True)

    def predict(self):
        if os.path.exists(PREDICTION_TARGET):
            self.ctx.log(
                "Prediction already exists."
                " If you want to re-run predict, just delete prediction file: " %
                PREDICTION_TARGET)
        else:
            Model(self.ctx, self.project).predict(
                PREDICTION_SOURCE, self.model_id, locally=True)

    def cleanup(self):
        self.project.delete()

def main():
    context = Context()
    try:
        app = ExampleApp(context)
        app.prepare_dataset()
        time.sleep(15)  # wait for redis is ready
        app.run_experiment()
        app.deploy()
        app.predict()
        app.cleanup()
    except Exception as e:
        import traceback; traceback.print_exc();
        context.log(
            "Example application execution has failed with error: '%s'" %
            str(e))


if __name__ == '__main__':
    main()