moderage-python 0.2.4

Mode Rage python client

Mode Rage client

What is ModeRage

ModeRage is a light-weight tool for storing experimental results and models. Experiments are referenced by their metacategory and id.

Meta Categories

Experiments in ModeRage have a meta-category, which basically defines the type of experiment. Think of meta-categories as an identifier for a project that may contain many experiments or datasets of the same type.

For example, when running many experiments with several sets of hyperparameters, those experiments can be saved into the same meta-category.

Ids

Once an experiment is saved it has an id. This can be used to load the experiment.

Configuration

ModeRage can be started in local or server mode.

Local

In local mode, ModeRage will save files locally to a ~/.moderage folder

Server

The ModeRage Server hosts experiments, data and metadata so it can be access from anywhere.

You can view it here (CURRENTLY UNDER DEVELOPMENT): Server

UI

The ModeRage UI communicates with the ModeRage server and allows browsing of experiments and data

You can view it here (CURRENTLY UNDER DEVELOPMENT): UI

Configuration file

Configuration in ModeRage is defined in a .mrconfig file. If no config file is created, ModeRage will automatically start in local Mode

Saving results

To save any number of files with some meta data you do the following:

1. Define a Meta object

mymeta = {
    'hyperparameter1': 100,
    'hyperparameter2': 200,
    'hyperparameter3': 0.7,
}

2. (Optional) Define any files you want to upload

myfiles = [
    {
        'filename': './path/to/myfile.csv',
        'caption': 'This is my file that contains my results'
    },
    {
        'filename': './path/to/mygraph.png',
        'caption': 'This is my file that contains my graph'
    },
    ...
]

4. (Optional) Reference any other experiments that this experiment is dependent on.

In many situations your experiment may rely on generated datasets or pre-trained models that also have many hyper-parameters. You can reference those parent experiments by adding them to the parent object

myparents = [
    {
        'id': [THE ID OF THE PARENT EXPERIEMENT],
        'metaCategory': 'generated_dataset'
    },
    {
        'id': [THE ID OF THE PARENT EXPERIEMENT],
        'metaCategory': 'pretrained_model'
    }
]

5. Call save

experiment = mr.save('category_name', mymeta, files=myfiles)

Loading results

Loading a saved experiment is simple, you just need to know the meta-category and the id of the experiment.

experiment = mr.load(id, meta_category)

Once the experiment is loaded, the meta information and files from the experiment can be accessed.

For example:

meta = experiment.meta
parents = experiment.parents
files = experiment.files

file = experiment.get_file('mygraph.png')