Bullet-Proof Hyperparameter Experiments with TensorFlow and Keras

Talos • Key Features • Examples • Install • Support • Docs • Issues • License • Download

Talos importantly improves ordinary TensorFlow (tf.keras) and Keras workflows by fully automating hyperparameter experiments and model evaluation. Talos exposes TensorFlow (tf.keras) and Keras functionality entirely and there is no new syntax or templates to learn.

The above animation illustrates how a minimal Sequntial model is modified for Talos

Talos

TL;DR Thousands of researchers have found Talos to importantly improve ordinary TensorFlow (tf.keras) and Keras workflows without taking away or hiding any of their power.

• Works with ANY Keras, TensorFlow (tf.keras) or PyTorch model
• Takes minutes to implement
• No new syntax to learn
• Adds zero new overhead to your workflow
• Bullet-proof results with no breaking bugs since 2019
• Comprehensive, up-to-date documentation

Talos is made for researchers, data scientists, and data engineers that want to remain in complete control of their TensorFlow (tf.keras) and Keras models, but are tired of mindless parameter hopping and confusing optimization solutions that add complexity instead of reducing it.

:wrench: Key Features

Within minutes, without learning any new syntax, Talos allows you to configure, perform, and evaluate hyperparameter experiments that yield state-of-the-art results across a wide range of prediction tasks. Talos provides the simplest and yet most powerful available method for hyperparameter optimization with TensorFlow (tf.keras) and Keras. Key features include:

• Single-line optimize-to-predict pipeline talos.Scan(x, y, model, params).predict(x_test, y_test)
• Automated hyperparameter optimization
• Model generalization evaluator
• Experiment analytics
• Pseudo, Quasi, and Quantum Random search options
• Grid search
• Probabilistic optimizers
• Single file custom optimization strategies
• Dynamically change optimization strategy during experiment
• Support for man-machine cooperative optimization strategy
• Model candidate generality evaluation
• Live training monitor
• Experiment analytics

Talos works on Linux, Mac OSX, and Windows systems and can be operated cpu, gpu, and multi-gpu systems.

:arrow_forward: Examples

Get the below code here. More examples further below.

The Simple example below is more than enough for starting to use Talos with any Keras model. Field Report has +4,400 claps on Medium because it's more entertaining.

Simple [1-2 mins]

Concise [~5 mins]

Comprehensive [~10 mins]

Field Report [~15 mins]

For more information on how Talos can help with your Keras, TensorFlow (tf.keras) and PyTorch workflow, visit the User Manual.

You may also want to check out a visualization of the Talos Hyperparameter Tuning workflow.

:floppy_disk: Install

Stable version:

pip install talos

Daily development version:

pip install git+https://github.com/autonomio/talos

:speech_balloon: How to get Support

:loudspeaker: Citations

If you use Talos for published work, please cite:

Autonomio Talos [Computer software]. (2024). Retrieved from http://github.com/autonomio/talos.

:page_with_curl: License

MIT License