Skip to main content

The minka package

Project description

About

Minka is ment to be a middleware between your optimization problem and a modular set of supporting tools to:

  • automatically perform hyperparameter optimization
  • store the results, parameter sets and custom data
  • compare different training runs
  • visualize the results and training processes

The aim of the project is to provide a simple as possible interface without the need to change your existing code.

One line to change it all

Could it be easier than just adding one line?

minka(yourOptimizationTask, 'configuration.json').opt(numberOfRuns)

Architecture overview

Usage

Installation

!pip install --upgrade PyYAML==4.2b4 optuna sacred python-telegram-bot dnspython loguru wandb pymongo

Preparation

Define your optimization problem

Minka uses a very simple combination of a configuration json file and the actual optimization task, represented as a single class.

optimization class

optimization template:

class yourOptimizationTask:
    def __init__(self):
        pass

    def prepare(self):
        pass

    def run(self, config):
        x = config['x']
        result = (x - 2) ** 2

        evalMetrics = {
            'error': result
        }

        logArrays = {
            'someValues': [1,2,3,4]
        }

        return result, evalMetrics, logArrays

    def cleanup(self):
        pass

parameter configuration

content of the configuration.json file:

{
    "comment": "parameter types: fix, categorical, discrete_uniform, int, loguniform, uniform",
    "parameters": {
        "batchSize": {"type": "fix","values": 256},
        "epochs": {"type": "fix","values": 75},
        "x": {"type": "categorical","values": [11,22,26]}
    }
}

Screenshots

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for minka-johann-haselberger, version 1.0.1
Filename, size File type Python version Upload date Hashes
Filename, size minka_johann_haselberger-1.0.1-py3-none-any.whl (4.3 kB) File type Wheel Python version py3 Upload date Hashes View hashes
Filename, size minka-johann-haselberger-1.0.1.tar.gz (4.0 kB) File type Source Python version None Upload date Hashes View hashes

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page