fantine / ml-hptuning

Machine learning and hyperparameter tuning framework

• Author: Fantine Huot

Getting started

Update the submodules

After cloning the repository, make sure to run the following commands to initialize and update the submodules.

git submodule init
git submodule update

Requirements

• TensorFlow

Folder structure

• bin: Scripts to run jobs.
• config: Configuration files.
• log: Log files.
• trainer: Machine learning model trainer.

Train a machine learning (ML) model

This repository provides a parameterized, modular framework for creating and running ML jobs.

Run a job

To train a machine learning model, use the following command:

bin/train.sh model_config dataset

• model_config: Name of ML model configuration to use. This should correspond to a configuration file named config/model_config.sh.
• dataset: Dataset identifier. Check the variables datapath, train_file, and eval_file in bin/train.sh to ensure that this maps to the correct input data.
• label: Optional label to add to the job name.

Set parameters for a job

Parameters for an ML job can be set by creating a corresponding configuration file: config/your_model_config.sh.

Create a new ML model architecture

• Create a new your_model.py file inside the trainer/model folder. Look at other models inside the folder for examples.
• Reference your new model in trainer/model/__init__.py.
• Set the model argument to your new model's name in your model configuration file config/your_model_config.sh.

Hyperparameter tuning

The hyperparameters are tuned using bayesian optimization.

Run a hyperparameter tuning job

To tune the hyperparameters for a machine learning model, use the following command:

bin/tunehp.sh model_config dataset

• model_config: Name of ML model configuration to use. This should correspond to a configuration file named config/model_config.sh.
• dataset: Dataset identifier. Check the variables datapath, train_file, and eval_file in bin/train.sh to ensure that this maps to the correct input data.

Define the domain for hyperparameter tuning

You can define the domain to explore for hyperparameter tuning by creating a corresponding configuration file: config/your_model_config_hptuning.yaml. Look at other hyperparameter tuning configuration files for examples.