Configuration

Giskard

Home Community

Search…

Get started

Guides

Installation & upgrade

Configuration

Upload your ML model & data

Evaluate your ML model

Test your ML model

Contribute to Giskard

Configuration

How to configure your Giskard application: Python version, Python version, email notifications and API tokens

Install additional Python libraries
If you need specific Python libraries that are not pre-installed in Giskard and that are available in PyPI, git, locally etc., you can install them manually in Giskard. 
To do so, please follow these steps:
1.Change the file giskard/giskard-ml-worker/ml-worker.dockerfile. To do that, here are the steps:
From your Giskard repo, type:
vim giskard-ml-worker/ml-worker.dockerfile
Below the line Add your python dependencies here, write the following line:  (press i to edit in the vim text editor).
RUN poetry add <YOUR LIBRARY> #With Poetry you can add libraries hosted locally, Github, PyPI, etc. See: https://python-poetry.org/docs/cli/#add etc
Save and quit the text editor (:wq in the vim editor)
2. Then rebuild your docker images using the following command lines
docker-compose -f docker-compose.dev.yml -f docker-compose.yml build ml-worker
docker-compose stop ml-worker && docker-compose up -d ml-worker
Install a new Python version
Giskard application is working on Python 3.7, if you use another Python version (3.9 or 3.10 for instance), please follow these steps:
1.Change Python version in giskard/giskard-ml-worker/ml-worker.dockerfile. To do that, here are the steps:
From your Giskard repo, type:
vim giskard-ml-worker/ml-worker.dockerfile
Replace Python3.7 in the first line by your Python version (press i to edit in the vim text editor). You can find the Python images according to your Python version in this dockerhub link​
Save and quit the text editor (:wq in the vim editor)
2. Then rebuild your docker images using the following command lines
docker-compose -f docker-compose.dev.yml -f docker-compose.yml build ml-worker
docker-compose stop ml-worker && docker-compose up -d ml-worker
Email notifications
Giskard can send email notifications about different events like inspector feedbacks or user invitation to the platform.
To configure this feature SMTP server credentials should be provided at startup through environment variables
Environment variable name
Purpose
GISKARD_MAIL_HOST
SMTP server address
GISKARD_MAIL_PORT
SMTP port, 587 by default
GISKARD_MAIL_USERNAME
​
GISKARD_MAIL_PASSWORD
​
GISKARD_MAIL_BASEURL
Publicly available Giskard instance URL
Backend JWT secret. Persistent user sessions.
Giskard authentication is based on JWT tokens. These tokens are issued by Giskard backend based on a JWT secret key.
To reinforce security there's no default value of the secret key, whenever the backend starts it generates one automatically. However it means that users' sessions are invalidated and they have to re-login. 
For production instances, it's preferred to keep user sessions alive no matter whether the server was rebooted or not. In this case, the JWT secret can be set from the outside by specifying GISKARD_JWT_SECRET environment variable. Its value should contain a base64 encoded bytes sequence of at least 128 bytes.
The following script can generate and store a secret key of 256 bytes:
# for zsh
echo export GISKARD_JWT_SECRET=`openssl rand -base64 256 | tr -d '\n'` >> ~/.zshrc
​
# for bash
echo export GISKARD_JWT_SECRET=`openssl rand -base64 256 | tr -d '\n'` >> ~/.bashrc
Troubleshooting​​
If you encounter any issues, join our Discord on our #support channel. Our community will help! 
​​​