update script to use s3

.github/workflows/print-loss.yml

@@ -22,14 +22,37 @@ jobs:
 
       - name: Install dependencies
         run: |
-          python -m pip install --upgrade pip
+          # python -m pip install --upgrade pip
           pip install -r requirements-dev.txt
 
       - name: Generate test data
         run: |
           printf '{"total_loss": 2.0}\n{"total_loss": 1.8023}\n{"total_loss": 1.52324}\n{"total_loss": 1.3234}' > test-log.jsonl
           ls -al
+      - name: Configure AWS credentials
+        uses: aws-actions/configure-aws-credentials@e3dd6a429d7300a6a4c196c26e071d42e0343502 # v4.0.2
+        with:
+          aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
+          aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
+          aws-region: ${{ vars.AWS_REGION }}
       - name: Print loss
         run: |
-          python scripts/create-loss-graph.py markdown --log-file test-log.jsonl --output-file 'results.md'
-          cat 'results.md' >> "${GITHUB_STEP_SUMMARY}"
+          python scripts/create-loss-graph.py image --log-file test-log.jsonl --output-file 'output.png'
+
+      - name: Try to upload to s3
+        run: |
+          echo "$(which aws)"
+          bucket_name='os-ci-loss-curve-test'
+          printf 'curious what this has: %s\n' "${{ github.event.pull_request.commits }}"
+          printf 'curious what this has: %s\n' "${{ github.event.pull_request.id }}"
+          printf 'curious what this has: %s\n' "${{ github.event.pull_request.head.sha }}"
+          base_path="pulls/${{ github.event.pull_request.number }}/curve.png"
+          s3_path="${bucket_name}/${base_path}"
+          aws s3 cp output.png "s3://${s3_path}"
+          
+          printf 'Test: us-east-2\n'
+
+          url="https://${bucket_name}.s3.${{vars.AWS_REGION }}.amazonaws.com/${base_path}"
+          printf 'url is: %s\n' "${url}"
+
+          printf '### Test performance:\n\n![Loss curve](%s)\n' "${url}" >> "${GITHUB_STEP_SUMMARY}"

requirements-dev.txt

@@ -1,6 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
--r requirements.txt
+# blot this out
+# -r requirements.txt
 
 pre-commit>=3.0.4,<5.0
 pylint>=2.16.2,<4.0

scripts/create-loss-graph.py

@@ -1,16 +1,19 @@
 # SPDX-License-Identifier: Apache-2.0
 # Standard
-from argparse import ArgumentParser, Namespace
-from base64 import b64encode
-from io import BytesIO
+from argparse import ArgumentParser
+from base64 import b64decode
+import os
 from pathlib import Path
 import json
+from typing import List
 
 # Third Party
 from matplotlib import pyplot as plt
 
+ENV_VAR_NAME = "LOSS_DATA"
 
-def create_b64_data(log_file: Path) -> str:
+
+def render_image(loss_data: List[float], outfile: Path) -> str:
     log_file = Path(args.log_file)
     if not log_file.exists():
         raise FileNotFoundError(f'log file "{args.log_file}" does not exist')
@@ -30,56 +33,83 @@ def create_b64_data(log_file: Path) -> str:
     plt.ylabel("Loss")
     plt.title("Training performance over fixed dataset")
 
-    buf = BytesIO()
-    plt.savefig(buf, format="png")
-    buf.seek(0)
-
-    imgb64 = b64encode(buf.read()).decode("utf-8")
-    return imgb64
-
-
-def create_md_file(b64_data: str, output_file: Path | None):
-    content = f"""## Training Performance\n
+    plt.savefig(outfile, format="png")
+
+
+def test_visibility():
+    envs = os.environ.keys()
+    print(envs)
+
+    print(
+        "===================================================================================================="
+    )
+    environ_name = "${ vars.AWS_REGION }"
+    print(f"Environment variable name: {environ_name}")
+    print(f'Environment variable value: {os.getenv(environ_name, "Not found")}')
+    print(
+        "===================================================================================================="
+    )
+    import subprocess
+
+    result = subprocess.run(
+        ["printf '%s' ${{ vars.AWS_REGION }}"], shell=True, capture_output=True
+    )
+    print(f"Result: {result.stdout.decode()}")
+
+
+def read_loss_data(src: str, log_file: Path | None = None) -> List[float]:
+    match src:
+        case "env":
+            data = os.getenv(ENV_VAR_NAME, None)
+            if not data:
+                raise ValueError(f"Environment variable {ENV_VAR_NAME} not set")
+            # decode the base64 data
+            data = b64decode(data)
+        case "file":
+            if not log_file:
+                raise ValueError("log_file must be provided when source is file")
+            if not log_file.exists():
+                raise FileNotFoundError(f"Log file {log_file} does not exist")
+            if log_file.is_dir():
+                raise ValueError(f"Log file {log_file} is a directory")
+            with open(log_file, "r") as f:
+                data = f.read()
+        case _:
+            raise ValueError(f"Unknown source: {src}")
 
-![Training Performance](data:image/png;base64,{b64_data})
-"""
-    if not output_file:
-        print(content)
-    else:
-        output_file.write_text(content, encoding="utf-8")
+    # select the loss data
+    contents = [json.loads(l) for l in data.splitlines()]
+    loss_data = [item["total_loss"] for item in contents if "total_loss" in item]
 
+    if not loss_data:
+        raise ValueError("Loss data is empty")
 
-def main(args: Namespace):
-    imgb64 = create_b64_data(args.log_file)
+    # ensure that the loss data is valid
+    if not all(isinstance(l, float) for l in loss_data):
+        raise ValueError("Loss data must be a list of floats")
 
-    output_file = Path(args.output_file) if args.output_file else None
-    if output_file:
-        output_file.write_text(imgb64, encoding="utf-8")
-    else:
-        # just print the file without including a newline, this way it can be piped
-        print(imgb64, end="")
+    return loss_data
 
 
 if __name__ == "__main__":
     parser = ArgumentParser()
     subparsers = parser.add_subparsers(dest="command", required=True)
 
     image_parser = subparsers.add_parser("image")
-    image_parser.add_argument("--log-file", type=str, required=True)
-    image_parser.add_argument("--output-file", type=str, default=None)
-
-    markdown_parser = subparsers.add_parser("markdown")
-    markdown_parser.add_argument("--log-file", type=str, required=True)
-    markdown_parser.add_argument("--output-file", type=str, default=None)
+    image_parser.add_argument(
+        "--source",
+        choices=["file", "env"],
+        default="file",
+        help="Source of the log file to read the loss data from. If file is selected, then we will read from the given file. If env is selected, we will read from the LOSS_DATA environment variable. If writing to env, the result should be a base64-encoded JSONL file.",
+    )
+    image_parser.add_argument("--log-file", type=str, default=None)
+    image_parser.add_argument("--output-file", type=str, required=True)
 
     args = parser.parse_args()
     match args.command:
         case "image":
-            print("creating image")
-            main(args)
-        case "markdown":
-            print("creating md file")
-            b64_data = create_b64_data(log_file=Path(args.log_file))
-            create_md_file(b64_data=b64_data, output_file=Path(args.output_file))
+            test_visibility()
+            loss_data = read_loss_data(src=args.source, log_file=Path(args.log_file))
+            render_image(loss_data=loss_data, outfile=Path(args.output_file))
         case _:
             raise ValueError(f"Unknown command: {args.command}")