How to Integrate Keras with Jira

Understanding the Integration Requirements

• Identify the objectives of integrating Keras with Jira. Typically, this involves reporting model training status or results to Jira for project management and documentation purposes.

 

• Ensure you have a running Jira instance and access to its API for creating and updating issues.

 

• Verify that you have a Python environment set up and Keras installed, along with necessary libraries like `requests` for making API calls.

Setup Jira API Access

• Generate an API token in Jira to authenticate API requests. Go to your account settings, then security, and create a new API token.

 

• Note down your Jira domain and project key. These will be used in API endpoints to interact with specific issues.

Prepare Your Environment

• Ensure Python is installed, preferably Python 3.6 or higher for compatibility with Keras.

 

• Install necessary libraries by running:

pip install keras requests

Set Up Keras Callback for Integration

• Create a custom Keras callback to send updates to Jira. This callback will trigger at defined points during the training process.

import requests
from keras.callbacks import Callback

class JiraCallback(Callback):
    def __init__(self, jira_url, api_token, project_key, issue_summary):
        super(JiraCallback, self).__init__()
        self.jira_url = jira_url
        self.headers = {
            'Authorization': f'Basic {api_token}',
            'Content-Type': 'application/json'
        }
        self.project_key = project_key
        self.issue_summary = issue_summary
    
    def on_epoch_end(self, epoch, logs=None):
        issue_data = {
            "fields": {
                "project": {
                    "key": self.project_key
                },
                "summary": f"{self.issue_summary} - Epoch {epoch + 1}",
                "description": f"Completed epoch {epoch + 1} with accuracy {logs['accuracy']:.4f} and loss {logs['loss']:.4f}.",
                "issuetype": {
                    "name": "Task"
                }
            }
        }
        response = requests.post(f"{self.jira_url}/rest/api/2/issue", headers=self.headers, json=issue_data)
        if response.status_code == 201:
            print(f"Created Jira issue for epoch {epoch + 1}")
        else:
            print(f"Failed to create Jira issue: {response.content}")

Integrate the Callback into Your Model Training

• Instantiate the `JiraCallback` with your Jira credentials and add it to the list of callbacks when fitting your model.

jira_callback = JiraCallback(
    jira_url='https://your-jira-domain.atlassian.net',
    api_token='your_encoded_api_token',
    project_key='PROJECT_KEY',
    issue_summary='Model Training Update'
)

model.fit(x_train, y_train, epochs=10, callbacks=[jira_callback])

Test Your Integration

• Run your model training and observe the console output for successful Jira issue creation messages.

 

• Log in to Jira and verify that issues are created and updated as expected with the training progress.

Troubleshoot Common Issues

• If issues are not created, check the API token and authorization headers for correctness.

 

• Ensure your Jira URL and endpoints are correct, especially if using a self-hosted Jira server.