Welcome to Jovian!¶

Jovian is a platform that helps data scientists and ML engineers:

Track & reproduce data science projects
Collaborate easily with friends/colleagues, and
Automate repetitive tasks in their day-to-day workflow.

Getting Started¶

Learn more about installing Jovian python library and some of the core features of Jovian.

Run this command in your terminal:

pip install jovian -q --upgrade

Installation¶

The jovian python library can be installed using the pip package manager. To install jovian via terminal or command line, run:

pip install jovian --upgrade

You can also install the jovian library directly within a Jupyter Notebook, by running the following command in a code cell:

!pip install jovian --upgrade

Caution

If you get a Permission denied error, try installing with sudo permission (on Linux/Mac).

$ sudo pip install jovian --upgrade

Another alternative is to try installing with the --user flag, but you’ll need to ensure that the target directory is added to your system PATH.

$ pip install jovian --upgrade --user

Once the installation is complete, you can start uploading Jupyter notebooks to Jovian.

Configuration (for Jovian Pro users only)

If you are a Jovian Pro user, run the following commands on the terminal (or command line) to connect the jovian library with your company’s internal Jovian Pro site:

jovian configure

You can also do this directly within a Jupyter notebook, by executing the following inside a code cell:

import jovian
jovian.configure()

The above command prompts for the following information:

Organization ID: The Organization ID provided by your company for authentication. E.g. if you are accessing Jovian Pro at https://mycompany.jvn.io , your organization ID is mycompany.
API key: You’ll get the API key when you’re logged in to your organization’s Jovian Pro site. By clicking on the API key button, the key will be copied to clipboard.

Note

You need to run jovian configure or jovian.configure() only once after installation. Your credentials are cached in the ~/.jovian directory on your computer. You can run jovian reset to clear this configuration.

You can learn more about Jovian Pro here, or start uploading Jupyter notebooks to Jovian in the next section.

Uploading Jupyter Notebooks to Jovian¶

Jovian allows you to upload and share Jupyter notebooks instantly with a single command, directly within Jupyter. Make sure you’ve completed the installation before reading further.

Uploading Notebooks¶

Step 1: Import jovian by running the following command within a Jupyter notebook.

import jovian

Step 2: After writing some code, running some experiments, training some models and plotting some charts, you can save and commit your Jupyter notebook.

jovian.commit()

When you run jovian.commit for the first time you’ll be asked to provide an API key, which you can get from your Jovian (or Jovian Pro) account.

Here’s what jovian.commit does:

It saves and uploads the Jupyter notebook to your Jovian (or Jovian Pro) account.
It captures and uploads the python virtual environment containing the list of libraries required to run your notebook.
It returns a link that you can use to view and share your notebook with friends or colleagues.

For more features of jovian.commit and API reference visit Commit.

Attention

In certain environments like JupyterLab and password protected notebooks, jovian may not be able to detect the notebook filename automatically. In such cases, pass the notebook’s name as the nb_filename argument to jovian.commit.

Benefits of Jovian¶

Easy sharing and collaboration: Just copy the link to share an uploaded notebook with your friends or colleages. Your notebooks are also visible on your profile page, unless you mark them Secret. You can also add collaborators and let others contribute to your project (learn more).

Cell-level comments and discussions: Jovian’s powerful commenting interface allows your team to discuss specific parts of a notebook with cell-level comment threads. Just hover over a cell and click the Comment button. You’ll receive an email when someone comments on your notebook, or replies to your comment.

End-to-end reproducibility: Jovian automatically captures Python libraries used in your notebook, so anyone (including you) can reproduce your work on any computer with a single command: jovian clone. You can also use the ‘Run’ dropdown on the Jovian notebook page to run your notebooks on free cloud GPU platforms like Google Colab, Kaggle Kernels and BinderHub.

This is just a small selection of features that Jovian offers. Continue reading by clicking the Next -> button to learn more, or use the sidebar to jump to a specific section.

Reproducing uploaded notebooks¶

A uploaded notebook on Jovian can be reproduced in any other machine. Follow the below procedure to do reproduce the notebook in your machine.

Clone¶

Visit the link of the uploaded notebook.
Click on the Clone button, which copies the clone cli command (along with the notebook_id) to the clipboard.
Paste the command in the terminal, in the directory where you want to clone the notebook project and then run the command.

The copied command will be of the the following format

jovian clone <notebook_id>

Install¶

Jovian captures the original python environment of the notebook, which make it easier to reproduce the notebook by installing all the required dependencies. The following commands uses conda to install all the required packages, make sure that conda is installed.

Once the notebook is cloned, it would have created a folder with the name of the notebook. Move to that directory.

cd jovian-demo

Then run

jovian install

The above command prompts for a virtual environment name where it will install all the required packages. By default it will have the original environment name in the square brackets, just click enter key to retain the name else specify the environment name.

In this way, Jovian seamlessly ensures the end-to-end reproducibility of your Jupyter notebooks across different operating systems.

Note

You have to own the notebook or have to be a collaborator to commit changes to the same project notebook. If not you can commit any changes made to your profile as a new notebook.

Pull¶

If changes are made after you have cloned the notebook by any of the collaborator. You can use pull to get all those changes.

Move to the cloned directory and run

jovian pull

Attention

Beware any uncommitted changes will be lost during the process of jovian.pull. When you pull the notebook it will be a duplicate of the latest version of the notebook on Jovian.

Notebook versioning and diffs¶

Version control¶

If you’re used to creating many duplicate versions of notebooks with slight modifications and long file names. Look no further, Jovian will be your version control for notebooks.

jovian.commit records all the versions under same notebook project. So, each change can be a version by author and collaborators which can be easily toggled in the website.

Note

You have to own the notebook or have to be a collaborator to commit changes to the same project notebook. If not you can commit any changes made to your profile as a new notebook.

View Differences¶

All the versions are comparable, you can view additions, deletions made among any 2 versions of the notebook. Hide/show common part of the code.

How to view the differences?

Commit different version and visit Jovian.
Click on Version drop down on the right top corner.
CLick on Compare Versions
Select any 2 versions with the use of check boxes and click on View Diff button.

There are more things to be compared, but first let’s add more content to the notebook to understand all the parameters that can be compared. Click on Next to follow through.

Attaching files and model outputs¶

As seen in the previous section by committing, source code and environment files are captured & uploaded. More files can be attached to the notebook such as files with helper code, output files/model checkpoints that the notebook is generating.

How to attach these files?¶

jovian.commit(files=[], artifacts=[])

What to include in the `files` argument?¶

The type of files which is required to run the notebook.

Helper code (.py)
Some input CSVs

What to include in the artifacts argument?¶

Any type of outputs that the notebook is generating.

Saved model or weights (.h5, .pkl, .pth)
Outputs, Submission CSVs
Images outputs

Where to search for the files after committing?¶

All the attached files are listed under Files Tab.

Files can be:

Renamed
Downloaded
Deleted
View Raw
Uploaded

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.

import jovian

Dataset¶

data = {
    'path': '/datasets/mnist',
    'description': '28x28 gray-scale images of handwritten digits'
}
jovian.log_dataset(data)

Hyperparameters¶

hyperparams = {
    'arch_name': 'cnn_1',
    'lr': .001
}
jovian.log_hyperparams(hyperparams)

Metrics¶

metrics = {
    'epoch': 1,
    'train_loss': .5,
    'val_loss': .3,
    'acc': .94
}
jovian.log_metrics(metrics)

Reset¶

If you’re not satisfied with some experiment and want to discard the previously recorded parameters to start a fresh experiment. Use

jovian.reset()

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.

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

Comparing and Analyzing experiments¶

Once you have more than 2 versions of a notebook, you will be able to use Compare Versions present in the Version dropdown on the top right corner.

Here you can observe all types of information about all of your versions.

Title
Time of Creation
Author
All the parameters logged under dataset.
All the parameters logged under hyperparameters.
All the parameters logged under metrics.
Notes (for author and collaborators add extra notes)

Sort¶

You can sort any column or a sub-column (For ex: accuracy or any other metric, date of creation etc.) by clicking on the column header.

Show, Hide and Reorder columns¶

You can create a custom view to analyse & compare your choice of parameters. Click on Configure button and then tick on the checkboxes to create a customized view. Click and drag the elements to reorder them based on your preference.

Add notes¶

You can add notes to summarize the experiment for reference or for collaborators to refer to.

View Diff between specific versions¶

Select any of the 2 versions by ticking the checkbox next to each version-row of the compare table which can be seen when you hover over any row. Click on View Diff button to view the additions and deletion made.

Archive/Delete versions¶

Select version/versions by ticking the checkbox of the row/rows. This enables both Archive and Delete ready for the respective actions.

Filter¶

By default all the archived versions are hidden, you can display them by enabling Show Archived in Filter dropdown.

Collaborating on Jovian projects¶

Jovian allows you to add collaborators to work with you on a ML Project.

How to add collaborators?¶

Click on Share button of the notebook and add them by their username or email id registered with Jovian.

This will allow the contributors to be able to commit changes to the same notebook project. The experiments by all the collaborators will also show up in the compare table. tab.

Comment on individual code cells¶

Users can comment on any code cells individually and maintain that thread to have specific discussion about a part of the source code with context.

Maintain secret notebooks¶

You may set a notebook as secret to hide it from publically being displayed on your profile. Only collaborators and users having the link to the notebook will be able to view it. You can find the option to Make Secret in the settings for each notebook.

Jovian Pro¶

Please contact us at hello@jvn.io

Commit¶

jovian.commit(secret=False, nb_filename=None, files=[], capture_env=True, env_type='conda', notebook_id=None, create_new=None, artifacts=[])[source]¶

Commits a Jupyter Notebook with its environment to Jovian.

Saves the checkpoint of the notebook, capture the required dependencies from the python environment and uploads the notebook, env file, additional files like scripts, csv etc. to https: // www.jvn.io . Capturing the python environment ensures that the notebook can be reproduced and executed easily using the ** {links to reprodue notebooks}.**

Parameters

secret (bool, optional) – Create a secret notebook on Jovian, which is only accessible via the link, and is not visible on the owner’s public profile. By default, committed notebooks are public and visible on the owner’s profile.
nb_filename (string, optional) – The filename of the jupyter notebook(including the .ipynb extension). This is detected automatically in most cases, but in certain environments like Jupyter Lab, the detection may fail and the filename needs to be provided using this argument.
files (array, optional) – Any additional scripts(.py files), CSVs that are required to run the notebook. These will be available in the files tab on Jovian .
capture_env (bool, optional) – If True, the Python environment(python version, libraries etc.) are captured and uploaded along with the notebook.
env_type (string, optional) – The type of environment to be captured. Allowed options are ‘conda’ and ‘pip’.
notebook_id (string, optional) – If you wish to update an existing notebook owned by you, you can use this argument to provide the base64 ID(present in the URL) of an notebook hosted on Jovian . In most cases, this argument is not required, and the library can automatically infer whether you are looking to update an existing notebook or create a new one.
create_new (bool, optional) – If set to True, doesn’t update the existing notebook on Jovian(if one is detected). Instead, it creates a new notebook when commit is called.
artifacts (array, optional) – Any outputs files or artifacts generated from the modeling processing. This can include model weights/checkpoints, generated CSVs, images etc.

Attention

Pass notebook’s name to nb_filename argument, in certain environments like Jupyter Lab and password protected notebooks sometimes it may fail to detect notebook automatically.

Log Dataset, Hyperparams & Metrics¶

jovian.log_dataset(data, verbose=True)[source]¶

Record dataset details for the current experiment

Parameters

data (dict) – A python dict or a array of dicts to be recorded as Dataset.
verbose (bool, optional) – By default it prints the acknowledgement, you can remove this by setting the argument to False.

Example

import jovian

data = {
    'path': '/datasets/mnist',
    'description': '28x28 images of handwritten digits (in grayscale)'
}
jovian.log_dataset(data)

jovian.log_hyperparams(data, verbose=True)[source]¶

Record hyperparameters for the current experiment

Parameters

data (dict) – A python dict or a array of dicts to be recorded as hyperparmeters.
verbose (bool, optional) – By default it prints the acknowledgement, you can remove this by setting the argument to False.

Example

import jovian

hyperparams = {
    'arch_name': 'cnn_1',
    'lr': .001
}
jovian.log_hyperparams(hyperparams)

jovian.log_metrics(data, verbose=True)[source]¶

Record metrics for the current experiment

Parameters

data (dict) – A python dict or a array of dicts to be recorded as metrics.
verbose (bool, optional) – By default it prints the acknowledgement, you can remove this by setting the argument to False.

Example

import jovian

metrics = {
    'epoch': 1,
    'train_loss': .5,
    'val_loss': .3,
    'acc': .94
}
jovian.log_metrics(metrics)

Command Line Commands¶

Initialize¶

Requests for a API Key for a new user, can find the key at Jovian. By clicking on API key button, key will be copied to the clipboard.

$ jovian init

Clone a Notebook¶

Clone a notebook form Jovian, by clicking on the Clone button of a notebook repo the whole clone command will be copied to the clipboard.

$ jovian clone {notebook_id}

Pull the latest Notebook¶

Pull the latest version of the notebook, use the command in a cloned repository or from a repository where you have committed to jovian.

$ jovian pull

Caution

Make sure the changes are committed if needed, pull overwrites the current notebook.

Install the required dependencies¶

Install all the dependencies required to the the cloned notebook, use the command in a cloned repository.

$ jovian install

Important

The above command prompts ` Please provide a name for the conda environment [{env_name}]: `

Press enter to install the dependencies to env_name (base env if the content of the square brackets is empty) else provide the env name in the prompt.

Version¶

Displays the current installed version of jovian library.

$ jovian version

Enable or Disable Jupyter Notebook Extension¶

By default, the jovian jupyter extension is enabled.

$ jovian enable-ext

$ jovian disable-ext

Note

The changes are observed when the webpage of the notebook is refreshed.

Fastai Callback¶

class jovian.callbacks.fastai.JovianFastaiCallback(learn: fastai.basic_train.Learner, arch_name: str)[source]¶

Fastai callback to automatically log hyperparameters and metrics.

Parameters

learn (Learner) – A learner object reference of your current model.
arch_name (string) – A name for the model you’re training.

Example

from jovian.callbacks.fastai_callback import FastaiCallback

jvn_cb = FastaiCallback(learn, 'res18')
learn.fit_one_cycle(5, callbacks = jvn_cb)

Tutorial

Visit this for a detailed example on using the keras callback, also visit the Records tab to see all the logs of that notebook logged by the callback.

Keras Callback¶

class jovian.callbacks.keras.JovianKerasCallback(reset_tracking=True, arch_name='', every_epoch=False, notify=False)[source]¶

Keras Callback to log hyperparameters and metrics during model training.

Parameters

reset_tracking (string, optional) – Will clear previously tracked hyperparameters & metrics, and start a fresh recording. Defaults to True.
arch_name (string, optional) – A name for the model you’re training.
every_epoch (bool, optional) – Whether to record losses & metrics for every epoch or just the final loss & metric. Defaults to False.
notify (bool, optional) – Whether to send notification on slack when the training ends. Defaults to False.

Example

from jovian.callbacks.keras import JovianKerasCallback

# To record logs of every epoch and to notify on slack
jvn_cb = JovianKerasCallback(arch_name='resnet18', every_epoch=True, notify=True)
model.fit(x_train, y_train, ...., callbacks=[jvn_cb])

Tutorial

Visit this for a detailed example on using the fastai callback, also visit the Records tab to see all the logs of that notebook logged by the callback.

Welcome to Jovian!¶

Getting Started¶

Installation¶

Uploading Jupyter Notebooks to Jovian¶

Uploading Notebooks¶

Benefits of Jovian¶

Reproducing uploaded notebooks¶

Clone¶

Install¶

Pull¶

Notebook versioning and diffs¶

Version control¶

View Differences¶

Attaching files and model outputs¶

How to attach these files?¶

What to include in the files argument?¶

What to include in the artifacts argument?¶

Where to search for the files after committing?¶

Tracking Datasets, Hyperparameters and Metrics¶

Dataset¶

Hyperparameters¶

Metrics¶

Reset¶

Comparing and Analyzing experiments¶

Sort¶

Show, Hide and Reorder columns¶

Add notes¶

View Diff between specific versions¶

Archive/Delete versions¶

Filter¶

Collaborating on Jovian projects¶

How to add collaborators?¶

Comment on individual code cells¶

Maintain secret notebooks¶

Jovian Pro¶

Commit¶

Log Dataset, Hyperparams & Metrics¶

Command Line Commands¶

Initialize¶

Clone a Notebook¶

Pull the latest Notebook¶

Install the required dependencies¶

Version¶

Enable or Disable Jupyter Notebook Extension¶

Fastai Callback¶

Keras Callback¶

Use Extension to Commit¶

Enable or Disable¶

What to include in the `files` argument?¶