fix: auto-create data drift issue (fix #5)

.github/workflows/data_drift_retrain.yml

@@ -7,6 +7,8 @@ on:
     workflows: ["Production Data Batching"]
     types:
       - completed
+  push:
+    branches: ijdoc/issue5
 
 jobs:
   drift_check:

batch/batch_data.py

@@ -65,7 +65,7 @@ def main(args):
         )
         artifact = run.log_artifact(artifact).wait()
         print(
-            f"New `{batch_type}` batch logged as `{artifact.source_name}`.\n"
+            f"Executed [run]({run.url}) to log new `{batch_type}` batch as `{artifact.source_name}`.\n"
             f"- Iteration: {iteration}\n"
             f"- Iteration Stride: {stride_days} day(s)\n"
             f"- Total length: {history_days} day(s)"

drift/check_drift.py

@@ -73,25 +73,24 @@
         artifact = run.log_artifact(artifact).wait()
         # Open a github issue asking for manual review
         issue_title = f"Data drift detected on {train_artifact.source_name}"
-        drifted_features = ", ".join(
-            [feature for feature, drift in drift_results.items() if drift]
-        )
         issue_body = (
-            f"Drift has been detected in the following features: {drifted_features}.\n\n"
-            f"Please review the [logged artifact](https://wandb.ai//{run.entity}/{run.project}/artifacts/{artifact.type}/{artifact.source_name}) "
-            f"and the [drift report]({report_url}) to determine if the training data should be updated.\n\n"
-            f"If approved, link the [logged artifact](https://wandb.ai//{run.entity}/{run.project}/artifacts/{artifact.type}/{artifact.source_name}) "
-            f"to the training Registry (`jdoc-org/wandb-registry-dataset/training`), otherwise, close this issue."
+            f"Data drift has been detected when comparing the registered training dataset with recent production data.\n\n"
+            f"Please review the [candidate artifact](https://wandb.ai/{run.entity}/{run.project}/artifacts/{artifact.type}/{artifact.source_name}) "
+            f"and the [drift report]({report_url}) to determine if the registered training data should be updated.\n\n"
+            f"To approve the new candidate after review, link it to [the training Dataset Registry](https://wandb.ai/registry/dataset?selectionPath=jdoc-org%2Fwandb-registry-dataset%2Ftraining&view=versions) at "
+            f"(`jdoc-org/wandb-registry-dataset/training`), otherwise close this issue."
         )
         issue_url = open_github_issue(issue_title, issue_body, labels=["drift", "data"])
         print(
-            f"Production batch `{prod_artifact.source_name}` was logged "
-            f"as candidate training data `{artifact.source_name}`. "
-            f"An [issue]({issue_url}) was created for manual review. "
+            f"Production batch `{prod_artifact.source_name}` has been logged "
+            f"as candidate to replace training data `{artifact.source_name}`. "
+            f"An [issue]({issue_url}) was created for manual review:\n"
         )
+        print(f"- [Data Drift Issue]({issue_url})")
     else:
         print("> No drift detected.\n")
 
+    print(f"- [W&B Run]({run.url})")
     print(f"- [Full data drift report]({report_url})")
 
     # Optionally the drift detection result in a parseable format.

drift/utils.py

@@ -174,19 +174,19 @@ def open_github_issue(issue_title, issue_body, labels=None):
     url = f"https://api.github.com/repos/{repo_owner}/{repo_name}/issues"
 
     headers = {
-        "Authorization": f"token {token}",
-        "Accept": "application/vnd.github.v3+json",
+        "Authorization": f"Bearer {token}",
+        "Accept": "application/vnd.github+json",
+        "X-GitHub-Api-Version": "2022-11-28",
     }
 
     data = {"title": issue_title, "body": issue_body, "labels": labels}
 
     response = requests.post(url, headers=headers, json=data)
     if response.status_code == 201:
         issue_url = response.json()["html_url"]
-        print(f"GitHub issue created: {issue_url}")
         return issue_url
     else:
-        print(f"Failed to create GitHub issue: {response.content}")
+        raise RuntimeError(f"Failed to create GitHub issue: {response.content}")
     return None
 
 
@@ -208,7 +208,7 @@ def get_github_repo_info():
             pattern = r"[email protected]:(.+)/(.+)\.git"
         elif remote_url.startswith("https://"):
             # HTTPS URL
-            pattern = r"https://github.com/(.+)/(.+)\.git"
+            pattern = r"https://github.com/(.+)/(.+)"
         else:
             # Other formats
             return None, None