Tracking Datasets, Hyperparameters and Metrics

Spreadsheets is one of the ways to track information & results of multiple ML experiments. However, using spreadsheets can be tiresome and non-intuitive without the context of the code.

Jovian makes its easy for anyone to track information about datasets, hyperparameters and metrics which are associated with each version of the your experiment in notebooks. Its also displays these information version-by-version of your notebook under single UI.

These information of a notebook are all added to Records Tab where you can toggle and view each version’s log.


data = {
    'path': '/datasets/mnist',
    'description': '28x28 gray-scale images of handwritten digits'


hyperparams = {
    'arch_name': 'cnn_1',
    'lr': .001


metrics = {
    'epoch': 1,
    'train_loss': .5,
    'val_loss': .3,
    'acc': .94
jovian track datasets, hyperparameters and metrics

The input to any of these can be a python dict . You can add custom parameters that are related to your experiment and have it record values manually, or automate it to record the values of a variable in a loop. Visit this page for these logging API reference.

We have callbacks for keras and fastai to automatically record hyperparams and metrics check it out.


If you’re not satisfied with some experiment and want to discard the current recorded logs before a commit. Use


Click Next to look at how to compare all of these information of all the versions.