openadapt-ml 0.2.0__py3-none-any.whl → 0.2.1__py3-none-any.whl

openadapt_ml/experiments/waa_demo/runner.py

@@ -26,7 +26,6 @@ import sys
 from typing import TYPE_CHECKING, Any
 
 from openadapt_ml.experiments.waa_demo.demos import (
-    DEMOS,
     format_demo_for_prompt,
     get_complete_demos,
     get_demo,
@@ -34,14 +33,16 @@ from openadapt_ml.experiments.waa_demo.demos import (
 )
 from openadapt_ml.experiments.waa_demo.tasks import (
     TASKS,
-    WATask,
-    get_manual_tasks,
     get_recorded_tasks,
     get_task,
 )
 
 if TYPE_CHECKING:
-    from openadapt_ml.benchmarks.base import BenchmarkAction, BenchmarkObservation, BenchmarkTask
+    from openadapt_evals import (
+        BenchmarkAction,
+        BenchmarkObservation,
+        BenchmarkTask,
+    )
 
 logger = logging.getLogger(__name__)
 
@@ -73,7 +74,9 @@ def cmd_list(args: argparse.Namespace) -> int:
     print()
     print("Tasks needing recorded demos on Windows:")
     for task in get_recorded_tasks():
-        print(f"  - #{list(TASKS.keys())[list(TASKS.values()).index(task)]}: {task.instruction}")
+        print(
+            f"  - #{list(TASKS.keys())[list(TASKS.values()).index(task)]}: {task.instruction}"
+        )
 
     return 0
 
@@ -122,7 +125,9 @@ def cmd_prompt(args: argparse.Namespace) -> int:
     else:
         print(f"Task: {task.instruction}")
         print()
-        print("Analyze the screenshot and provide the next action to complete this task.")
+        print(
+            "Analyze the screenshot and provide the next action to complete this task."
+        )
         if demo and "[PLACEHOLDER" in demo:
             print()
             print("[Note: Demo not available - this would be zero-shot]")
@@ -208,6 +213,7 @@ Think step by step, then output the action on a new line starting with "ACTION:"
         """Lazily initialize the API adapter."""
         if self._adapter is None:
             from openadapt_ml.models.api_adapter import ApiVLMAdapter
+
             self._adapter = ApiVLMAdapter(
                 provider=self.provider,
                 api_key=self.api_key,
@@ -261,7 +267,7 @@ Think step by step, then output the action on a new line starting with "ACTION:"
         Returns:
             BenchmarkAction parsed from VLM response
         """
-        from openadapt_ml.benchmarks.base import BenchmarkAction
+        from openadapt_evals import BenchmarkAction
 
         adapter = self._get_adapter()
 
@@ -325,7 +331,9 @@ Think step by step, then output the action on a new line starting with "ACTION:"
             history_str = self._format_history(history)
             content_parts.append(f"Previous actions:\n{history_str}")
 
-        content_parts.append("\nAnalyze the current screenshot and provide the next action.")
+        content_parts.append(
+            "\nAnalyze the current screenshot and provide the next action."
+        )
 
         sample: dict[str, Any] = {
             "messages": [
@@ -401,7 +409,7 @@ Think step by step, then output the action on a new line starting with "ACTION:"
         Uses the same parsing logic as APIBenchmarkAgent.
         """
         import re
-        from openadapt_ml.benchmarks.base import BenchmarkAction
+        from openadapt_evals import BenchmarkAction
 
         raw_action = {"response": response}
 
@@ -457,7 +465,9 @@ Think step by step, then output the action on a new line starting with "ACTION:"
             r"TYPE\s*\(\s*[\"'](.+?)[\"']\s*\)", action_line, re.IGNORECASE
         )
         if type_match:
-            return BenchmarkAction(type="type", text=type_match.group(1), raw_action=raw_action)
+            return BenchmarkAction(
+                type="type", text=type_match.group(1), raw_action=raw_action
+            )
 
         # Parse KEY
         key_match = re.match(r"KEY\s*\(\s*(.+?)\s*\)", action_line, re.IGNORECASE)
@@ -502,11 +512,12 @@ def cmd_run(args: argparse.Namespace) -> int:
     This integrates with the benchmarks infrastructure to run either
     zero-shot or demo-conditioned evaluation on WAA tasks.
     """
-    from openadapt_ml.benchmarks import (
+    from openadapt_evals import (
+        EvaluationConfig,
         WAAMockAdapter,
         compute_metrics,
+        evaluate_agent_on_benchmark,
     )
-    from openadapt_ml.benchmarks.runner import EvaluationConfig, evaluate_agent_on_benchmark
 
     print("WAA Demo-Conditioned Experiment Runner")
     print("=" * 80)
@@ -539,7 +550,7 @@ def cmd_run(args: argparse.Namespace) -> int:
         print(f"Running {len(task_ids)} tasks with complete demos")
 
     # Check for mock mode or real WAA
-    use_mock = getattr(args, 'mock', False)
+    use_mock = getattr(args, "mock", False)
 
     if use_mock:
         print("Using mock adapter (no Windows required)")
@@ -599,7 +610,11 @@ def cmd_run(args: argparse.Namespace) -> int:
     except Exception as e:
         print(f"Error during evaluation: {e}")
         if "API key" in str(e) or "api_key" in str(e).lower():
-            key_name = "ANTHROPIC_API_KEY" if args.provider == "anthropic" else "OPENAI_API_KEY"
+            key_name = (
+                "ANTHROPIC_API_KEY"
+                if args.provider == "anthropic"
+                else "OPENAI_API_KEY"
+            )
             print(f"\nMake sure {key_name} is set in your environment or .env file.")
         return 1
 

openadapt_ml/export/parquet.py

@@ -7,7 +7,6 @@ Episode JSON remains the canonical representation.
 from __future__ import annotations
 
 import json
-from pathlib import Path
 from typing import TYPE_CHECKING
 
 if TYPE_CHECKING:
@@ -73,7 +72,11 @@ def to_parquet(
             # Extract action type value (enum -> string)
             action_type = None
             if step.action:
-                action_type = step.action.type.value if hasattr(step.action.type, 'value') else step.action.type
+                action_type = (
+                    step.action.type.value
+                    if hasattr(step.action.type, "value")
+                    else step.action.type
+                )
 
             row = {
                 "episode_id": episode.episode_id,
@@ -84,8 +87,12 @@ def to_parquet(
                 "action_type": action_type,
                 "x": x,
                 "y": y,
-                "end_x": step.action.normalized_end[0] if step.action and step.action.normalized_end else None,
-                "end_y": step.action.normalized_end[1] if step.action and step.action.normalized_end else None,
+                "end_x": step.action.normalized_end[0]
+                if step.action and step.action.normalized_end
+                else None,
+                "end_y": step.action.normalized_end[1]
+                if step.action and step.action.normalized_end
+                else None,
                 "text": getattr(step.action, "text", None) if step.action else None,
                 "key": getattr(step.action, "key", None) if step.action else None,
                 "scroll_direction": (
@@ -131,33 +138,37 @@ def _write_summary(episodes: list[Episode], output_path: str) -> None:
     for episode in episodes:
         first_t = episode.steps[0].timestamp if episode.steps else None
         last_t = episode.steps[-1].timestamp if episode.steps else None
-        duration = (last_t - first_t) if first_t is not None and last_t is not None else None
+        duration = (
+            (last_t - first_t) if first_t is not None and last_t is not None else None
+        )
 
         # Extract action type values (enum -> string)
         first_action_type = None
         last_action_type = None
         if episode.steps and episode.steps[0].action:
             t = episode.steps[0].action.type
-            first_action_type = t.value if hasattr(t, 'value') else t
+            first_action_type = t.value if hasattr(t, "value") else t
         if episode.steps and episode.steps[-1].action:
             t = episode.steps[-1].action.type
-            last_action_type = t.value if hasattr(t, 'value') else t
-
-        summary_rows.append({
-            "episode_id": episode.episode_id,
-            "instruction": episode.instruction,
-            "task_id": getattr(episode, "task_id", None),
-            "step_count": len(episode.steps),
-            "duration": duration,
-            "success": getattr(episode, "success", None),
-            "first_action_type": first_action_type,
-            "last_action_type": last_action_type,
-            "metadata": (
-                json.dumps(episode.metadata)
-                if hasattr(episode, "metadata") and episode.metadata
-                else None
-            ),
-        })
+            last_action_type = t.value if hasattr(t, "value") else t
+
+        summary_rows.append(
+            {
+                "episode_id": episode.episode_id,
+                "instruction": episode.instruction,
+                "task_id": getattr(episode, "task_id", None),
+                "step_count": len(episode.steps),
+                "duration": duration,
+                "success": getattr(episode, "success", None),
+                "first_action_type": first_action_type,
+                "last_action_type": last_action_type,
+                "metadata": (
+                    json.dumps(episode.metadata)
+                    if hasattr(episode, "metadata") and episode.metadata
+                    else None
+                ),
+            }
+        )
 
     summary_table = pa.Table.from_pylist(summary_rows)
     summary_path = str(output_path).replace(".parquet", "_summary.parquet")
@@ -255,7 +266,8 @@ def from_parquet(parquet_path: str) -> list[Episode]:
 
         episode = Episode(
             episode_id=str(episode_id),
-            instruction=group.iloc[0].get("instruction") or group.iloc[0].get("goal", ""),
+            instruction=group.iloc[0].get("instruction")
+            or group.iloc[0].get("goal", ""),
             steps=steps,
             task_id=group.iloc[0].get("task_id"),
             metadata=metadata,

openadapt_ml/grounding/detector.py

@@ -20,7 +20,7 @@ from openadapt_ml.config import settings
 from openadapt_ml.grounding.base import GroundingModule, RegionCandidate
 
 if TYPE_CHECKING:
-    from PIL import Image, ImageDraw, ImageFont
+    from PIL import Image
 
 
 class GeminiGrounder(GroundingModule):
@@ -104,7 +104,7 @@ class GeminiGrounder(GroundingModule):
 
         # Try to parse JSON from the response
         # Look for JSON array or object in the response
-        json_match = re.search(r'\[[\s\S]*\]|\{[\s\S]*\}', response_text)
+        json_match = re.search(r"\[[\s\S]*\]|\{[\s\S]*\}", response_text)
         if not json_match:
             return candidates
 
@@ -340,11 +340,11 @@ Example output format:
         response_text = response.text
 
         # Try to extract JSON array from response
-        json_match = re.search(r'\[[\s\S]*\]', response_text)
+        json_match = re.search(r"\[[\s\S]*\]", response_text)
         if not json_match:
             # Maybe it's just a plain array
-            if response_text.strip().startswith('['):
-                json_match = re.match(r'.*', response_text)
+            if response_text.strip().startswith("["):
+                json_match = re.match(r".*", response_text)
             else:
                 return []
 
@@ -369,13 +369,18 @@ Example output format:
                         max(0, min(1, y2 / screenshot.height)),
                     ]
 
-                normalized_elements.append({
-                    "id": elem.get("id", len(normalized_elements) + 1),
-                    "label": elem.get("label", f"Element {elem.get('id', len(normalized_elements) + 1)}"),
-                    "bbox": norm_bbox,
-                    "type": elem.get("type", "other"),
-                    "text": elem.get("text", ""),
-                })
+                normalized_elements.append(
+                    {
+                        "id": elem.get("id", len(normalized_elements) + 1),
+                        "label": elem.get(
+                            "label",
+                            f"Element {elem.get('id', len(normalized_elements) + 1)}",
+                        ),
+                        "bbox": norm_bbox,
+                        "type": elem.get("type", "other"),
+                        "text": elem.get("text", ""),
+                    }
+                )
 
         return normalized_elements
 
@@ -549,8 +554,7 @@ class DetectorGrounder(GroundingModule):
             self._backend = GeminiGrounder(**kwargs)
         elif backend == "omniparser":
             raise NotImplementedError(
-                "OmniParser backend not yet implemented. "
-                "Use backend='gemini' for now."
+                "OmniParser backend not yet implemented. Use backend='gemini' for now."
             )
         else:
             raise ValueError(f"Unknown backend: {backend}")

openadapt_ml/ingest/__init__.py

@@ -27,16 +27,18 @@ __all__ = [
 
 # Conditionally export capture functions if openadapt-capture is installed
 try:
-    from openadapt_ml.ingest.capture import (
+    from openadapt_ml.ingest.capture import (  # noqa: F401
         capture_to_episode,
         capture_to_session,
         load_captures_as_sessions,
     )
 
-    __all__.extend([
-        "capture_to_episode",
-        "capture_to_session",
-        "load_captures_as_sessions",
-    ])
+    __all__.extend(
+        [
+            "capture_to_episode",
+            "capture_to_session",
+            "load_captures_as_sessions",
+        ]
+    )
 except ImportError:
     pass

openadapt_ml/ingest/capture.py

@@ -6,7 +6,6 @@ and convert them to the Episode/Step format used by openadapt-ml for training.
 
 from __future__ import annotations
 
-import uuid
 from pathlib import Path
 from typing import TYPE_CHECKING
 
@@ -101,7 +100,7 @@ def capture_to_episode(
     """
     try:
         from openadapt_capture import Capture
-        from openadapt_capture.events import (
+        from openadapt_capture.events import (  # noqa: F401
             EventType,
             KeyTypeEvent,
             MouseClickEvent,
@@ -135,7 +134,9 @@ def capture_to_episode(
             dir_name = capture_path.name
             if dir_name and dir_name != "capture":
                 # Convert kebab-case/snake_case to readable text
-                instruction = dir_name.replace("-", " ").replace("_", " ").strip().capitalize()
+                instruction = (
+                    dir_name.replace("-", " ").replace("_", " ").strip().capitalize()
+                )
             else:
                 instruction = "Complete the recorded workflow"
 
@@ -155,9 +156,7 @@ def capture_to_episode(
         screenshot_path = _save_screenshot(screenshot, output_dir, episode_id, idx)
 
         # Normalize coordinates
-        norm_coords = _normalize_coords(
-            action.x, action.y, screen_width, screen_height
-        )
+        norm_coords = _normalize_coords(action.x, action.y, screen_width, screen_height)
 
         # Map event type to openadapt-ml ActionType
         event_type = action.type
@@ -174,15 +173,15 @@ def capture_to_episode(
         if isinstance(action.event, MouseDragEvent):
             end_x = action.event.x + action.event.dx
             end_y = action.event.y + action.event.dy
-            norm_end = _normalize_coords(
-                end_x, end_y, screen_width, screen_height
+            norm_end = _normalize_coords(end_x, end_y, screen_width, screen_height)
+            ml_action = ml_action.model_copy(
+                update={
+                    "normalized_end": norm_end,
+                    "raw": {
+                        "button": action.event.button,
+                    },
+                }
             )
-            ml_action = ml_action.model_copy(update={
-                "normalized_end": norm_end,
-                "raw": {
-                    "button": action.event.button,
-                },
-            })
 
         # Handle scroll events
         if isinstance(action.event, MouseScrollEvent):
@@ -197,13 +196,15 @@ def capture_to_episode(
             elif action.event.dx < 0:
                 scroll_direction = "left"
 
-            ml_action = ml_action.model_copy(update={
-                "scroll_direction": scroll_direction,
-                "raw": {
-                    "dx": action.event.dx,
-                    "dy": action.event.dy,
-                },
-            })
+            ml_action = ml_action.model_copy(
+                update={
+                    "scroll_direction": scroll_direction,
+                    "raw": {
+                        "dx": action.event.dx,
+                        "dy": action.event.dy,
+                    },
+                }
+            )
 
         # Handle keyboard events - include key names for special keys
         if action.keys:
@@ -227,7 +228,9 @@ def capture_to_episode(
         last_step = steps[-1]
         done_step = Step(
             step_index=len(steps),
-            observation=Observation(screenshot_path=last_step.observation.screenshot_path),
+            observation=Observation(
+                screenshot_path=last_step.observation.screenshot_path
+            ),
             action=Action(type=ActionType.DONE),
             reasoning="Workflow complete.",
             timestamp=(last_step.timestamp or 0) + 0.1,

openadapt_ml/ingest/loader.py

@@ -8,9 +8,8 @@ from __future__ import annotations
 
 import json
 from pathlib import Path
-from typing import Any, Dict, List, Optional, Union
+from typing import Any, Dict, List, Union
 
-from pydantic import ValidationError
 
 from openadapt_ml.schema import Action, ActionType, Episode, Observation, Step
 
@@ -143,7 +142,8 @@ def _dict_to_episode(data: Dict[str, Any], validate: bool = True) -> Episode:
         # Parse observation
         obs_data = step_data.get("observation", {})
         observation = Observation(
-            screenshot_path=obs_data.get("screenshot_path") or obs_data.get("image_path"),
+            screenshot_path=obs_data.get("screenshot_path")
+            or obs_data.get("image_path"),
             raw=obs_data.get("raw") or obs_data.get("meta"),
             a11y_tree=obs_data.get("a11y_tree") or obs_data.get("accessibility_tree"),
             dom=obs_data.get("dom") or obs_data.get("dom_html"),
@@ -169,7 +169,10 @@ def _dict_to_episode(data: Dict[str, Any], validate: bool = True) -> Episode:
         normalized_end = None
         if action_data.get("normalized_end"):
             normalized_end = tuple(action_data["normalized_end"])
-        elif action_data.get("end_x") is not None and action_data.get("end_y") is not None:
+        elif (
+            action_data.get("end_x") is not None
+            and action_data.get("end_y") is not None
+        ):
             normalized_end = (action_data["end_x"], action_data["end_y"])
 
         action = Action(