synth-ai 0.2.14__py3-none-any.whl → 0.4.4__py3-none-any.whl

synth_ai/sdk/learning/context_learning_client.py

@@ -0,0 +1,531 @@
+"""Client for interacting with context learning jobs.
+
+Context Learning optimizes environment setup scripts (pre-flight/post-flight bash)
+for terminal/coding agents. This is "prompt optimization" for terminal agents.
+
+Example usage:
+    from synth_ai.sdk.learning import ContextLearningClient, ContextLearningJobConfig
+
+    client = ContextLearningClient("http://localhost:8000", "your-api-key")
+    
+    # Create a job from config
+    config = ContextLearningJobConfig.from_dict({
+        "task_app_url": "http://localhost:8102",
+        "evaluation_seeds": [0, 1, 2, 3, 4],
+        "environment": {
+            "preflight_script": "#!/bin/bash\\necho 'Setting up...'",
+        },
+        "algorithm_config": {
+            "initial_population_size": 10,
+            "num_generations": 3,
+        },
+    })
+    
+    job_id = await client.create_job(config)
+    
+    # Stream events
+    async for event in client.stream_events(job_id):
+        print(f"{event.event_type}: {event.message}")
+    
+    # Get best script when done
+    result = await client.get_best_script(job_id)
+    print(result.preflight_script)
+"""
+
+from __future__ import annotations
+
+import asyncio
+from typing import Any, AsyncIterator, Dict, List, Optional
+
+from synth_ai.core._utils.http import AsyncHttpClient
+
+from .context_learning_types import (
+    BestScriptResult,
+    ContextLearningEvent,
+    ContextLearningJobConfig,
+    ContextLearningJobStatus,
+    ContextLearningMetric,
+    ContextLearningResults,
+)
+
+
+def _validate_job_id(job_id: str) -> None:
+    """Validate that job_id has the expected context learning format.
+    
+    Args:
+        job_id: Job ID to validate
+        
+    Raises:
+        ValueError: If job_id doesn't start with 'cl_job_'
+    """
+    if not job_id.startswith("cl_job_"):
+        raise ValueError(
+            f"Invalid context learning job ID format: {job_id!r}. "
+            f"Expected format: 'cl_job_<identifier>' (e.g., 'cl_job_abc123def456')"
+        )
+
+
+class ContextLearningClient:
+    """Async client for interacting with context learning jobs.
+    
+    This client provides a Pythonic interface for:
+    - Creating and managing context learning jobs
+    - Streaming events and metrics
+    - Retrieving best scripts and results
+    
+    Example:
+        >>> client = ContextLearningClient("http://localhost:8000", "your-api-key")
+        >>> job_id = await client.create_job(config)
+        >>> status = await client.get_job(job_id)
+        >>> print(status.status)  # "running"
+    """
+
+    def __init__(self, base_url: str, api_key: str, *, timeout: float = 60.0) -> None:
+        """Initialize the context learning client.
+        
+        Args:
+            base_url: Base URL of the backend API (e.g., "http://localhost:8000")
+            api_key: API key for authentication
+            timeout: Request timeout in seconds (default: 60s for long operations)
+        """
+        self._base_url = base_url.rstrip("/")
+        self._api_key = api_key
+        self._timeout = timeout
+
+    # -------------------------------------------------------------------------
+    # Job Management
+    # -------------------------------------------------------------------------
+
+    async def create_job(
+        self,
+        config: ContextLearningJobConfig | Dict[str, Any],
+    ) -> str:
+        """Create a new context learning job.
+        
+        Args:
+            config: Job configuration (ContextLearningJobConfig or dict)
+            
+        Returns:
+            Job ID (e.g., "cl_job_abc123def456")
+            
+        Raises:
+            RuntimeError: If job creation fails
+        """
+        if isinstance(config, dict):
+            config = ContextLearningJobConfig.from_dict(config)
+        
+        payload = config.to_dict()
+        
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            result = await http.post_json("/api/context-learning/jobs", json=payload)
+        
+        job_id = result.get("job_id")
+        if not job_id:
+            raise RuntimeError(f"Job creation failed: response missing job_id. Response: {result}")
+        
+        return job_id
+
+    async def get_job(self, job_id: str) -> ContextLearningJobStatus:
+        """Get job details and status.
+        
+        Args:
+            job_id: Job ID
+            
+        Returns:
+            Job status with details
+            
+        Raises:
+            ValueError: If job_id format is invalid
+        """
+        _validate_job_id(job_id)
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            result = await http.get(f"/api/context-learning/jobs/{job_id}")
+        return ContextLearningJobStatus.from_dict(result)
+
+    async def list_jobs(self, *, limit: int = 50) -> List[ContextLearningJobStatus]:
+        """List context learning jobs.
+        
+        Args:
+            limit: Maximum number of jobs to return
+            
+        Returns:
+            List of job statuses, sorted by created_at descending
+        """
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            result = await http.get("/api/context-learning/jobs", params={"limit": limit})
+        
+        if isinstance(result, list):
+            return [ContextLearningJobStatus.from_dict(j) for j in result]
+        return []
+
+    async def start_job(self, job_id: str) -> ContextLearningJobStatus:
+        """Start a pending job.
+        
+        Args:
+            job_id: Job ID
+            
+        Returns:
+            Updated job status
+            
+        Raises:
+            ValueError: If job_id format is invalid
+            RuntimeError: If job cannot be started
+        """
+        _validate_job_id(job_id)
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            result = await http.post_json(f"/api/context-learning/jobs/{job_id}/start", json={})
+        return ContextLearningJobStatus.from_dict(result)
+
+    async def cancel_job(self, job_id: str) -> ContextLearningJobStatus:
+        """Cancel a running job.
+        
+        Args:
+            job_id: Job ID
+            
+        Returns:
+            Updated job status
+            
+        Raises:
+            ValueError: If job_id format is invalid
+            RuntimeError: If job cannot be cancelled
+        """
+        _validate_job_id(job_id)
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            result = await http.post_json(f"/api/context-learning/jobs/{job_id}/cancel", json={})
+        return ContextLearningJobStatus.from_dict(result)
+
+    # -------------------------------------------------------------------------
+    # Events and Metrics
+    # -------------------------------------------------------------------------
+
+    async def get_events(
+        self, job_id: str, *, limit: int = 100
+    ) -> List[ContextLearningEvent]:
+        """Get events for a job.
+        
+        Args:
+            job_id: Job ID
+            limit: Maximum number of events to return
+            
+        Returns:
+            List of events, most recent first
+            
+        Raises:
+            ValueError: If job_id format is invalid
+        """
+        _validate_job_id(job_id)
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            result = await http.get(
+                f"/api/context-learning/jobs/{job_id}/events",
+                params={"limit": limit},
+            )
+        
+        if isinstance(result, list):
+            return [ContextLearningEvent.from_dict(e) for e in result]
+        return []
+
+    async def get_metrics(
+        self, job_id: str, *, name: Optional[str] = None, limit: int = 500
+    ) -> List[ContextLearningMetric]:
+        """Get metrics for a job.
+        
+        Args:
+            job_id: Job ID
+            name: Optional metric name filter
+            limit: Maximum number of metrics to return
+            
+        Returns:
+            List of metrics
+            
+        Raises:
+            ValueError: If job_id format is invalid
+        """
+        _validate_job_id(job_id)
+        params: Dict[str, Any] = {"limit": limit}
+        if name:
+            params["name"] = name
+        
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            result = await http.get(
+                f"/api/context-learning/jobs/{job_id}/metrics",
+                params=params,
+            )
+        
+        points = result.get("points", []) if isinstance(result, dict) else []
+        return [ContextLearningMetric.from_dict(p) for p in points]
+
+    async def stream_events(
+        self,
+        job_id: str,
+        *,
+        poll_interval: float = 2.0,
+        timeout: Optional[float] = None,
+    ) -> AsyncIterator[ContextLearningEvent]:
+        """Stream events from a job until it reaches a terminal state.
+        
+        This is a polling-based implementation that yields new events as they arrive.
+        
+        Args:
+            job_id: Job ID
+            poll_interval: Seconds between polls
+            timeout: Maximum seconds to stream (None = no timeout)
+            
+        Yields:
+            Events as they occur
+            
+        Raises:
+            ValueError: If job_id format is invalid
+            asyncio.TimeoutError: If timeout is reached
+        """
+        _validate_job_id(job_id)
+        
+        seen_seqs: set[int] = set()
+        start_time = asyncio.get_event_loop().time()
+        
+        while True:
+            # Check timeout
+            if timeout is not None:
+                elapsed = asyncio.get_event_loop().time() - start_time
+                if elapsed >= timeout:
+                    raise asyncio.TimeoutError(f"Stream timeout after {timeout}s")
+            
+            # Get job status
+            status = await self.get_job(job_id)
+            
+            # Get events
+            events = await self.get_events(job_id, limit=500)
+            
+            # Yield new events
+            for event in events:
+                seq = event.seq
+                if seq is not None and seq not in seen_seqs:
+                    seen_seqs.add(seq)
+                    yield event
+            
+            # Check if done
+            if status.is_terminal:
+                break
+            
+            # Wait before next poll
+            await asyncio.sleep(poll_interval)
+
+    # -------------------------------------------------------------------------
+    # Results
+    # -------------------------------------------------------------------------
+
+    async def get_best_script(self, job_id: str) -> BestScriptResult:
+        """Get the best performing pre-flight script.
+        
+        Args:
+            job_id: Job ID
+            
+        Returns:
+            Best script result with score and metadata
+            
+        Raises:
+            ValueError: If job_id format is invalid
+            RuntimeError: If job is not completed or no best script available
+        """
+        _validate_job_id(job_id)
+        async with AsyncHttpClient(self._base_url, self._api_key, timeout=self._timeout) as http:
+            result = await http.get(f"/api/context-learning/jobs/{job_id}/best-script")
+        return BestScriptResult.from_dict(result)
+
+    async def get_results(self, job_id: str) -> ContextLearningResults:
+        """Get complete results from a job.
+        
+        Aggregates status, events, and best script into a single result object.
+        
+        Args:
+            job_id: Job ID
+            
+        Returns:
+            Complete results including status, events, and best script
+            
+        Raises:
+            ValueError: If job_id format is invalid
+        """
+        _validate_job_id(job_id)
+        
+        status = await self.get_job(job_id)
+        events = await self.get_events(job_id, limit=1000)
+        
+        best_script: Optional[BestScriptResult] = None
+        if status.is_successful:
+            try:
+                best_script = await self.get_best_script(job_id)
+            except Exception:
+                pass
+        
+        return ContextLearningResults.from_status_and_events(status, events, best_script)
+
+    # -------------------------------------------------------------------------
+    # Convenience Methods
+    # -------------------------------------------------------------------------
+
+    async def wait_for_completion(
+        self,
+        job_id: str,
+        *,
+        poll_interval: float = 5.0,
+        timeout: Optional[float] = None,
+        on_event: Optional[callable] = None,
+    ) -> ContextLearningJobStatus:
+        """Wait for a job to complete.
+        
+        Args:
+            job_id: Job ID
+            poll_interval: Seconds between status checks
+            timeout: Maximum seconds to wait (None = no timeout)
+            on_event: Optional callback for each new event
+            
+        Returns:
+            Final job status
+            
+        Raises:
+            ValueError: If job_id format is invalid
+            asyncio.TimeoutError: If timeout is reached
+        """
+        async for event in self.stream_events(job_id, poll_interval=poll_interval, timeout=timeout):
+            if on_event:
+                try:
+                    on_event(event)
+                except Exception:
+                    pass
+        
+        return await self.get_job(job_id)
+
+    async def run_job(
+        self,
+        config: ContextLearningJobConfig | Dict[str, Any],
+        *,
+        poll_interval: float = 5.0,
+        timeout: Optional[float] = None,
+        on_event: Optional[callable] = None,
+    ) -> ContextLearningResults:
+        """Create a job, wait for completion, and return results.
+        
+        This is the main entry point for running context learning jobs.
+        
+        Args:
+            config: Job configuration
+            poll_interval: Seconds between status checks
+            timeout: Maximum seconds to wait (None = no timeout)
+            on_event: Optional callback for each new event
+            
+        Returns:
+            Complete results including best script
+            
+        Example:
+            >>> config = ContextLearningJobConfig.from_dict({
+            ...     "task_app_url": "http://localhost:8102",
+            ...     "evaluation_seeds": [0, 1, 2],
+            ... })
+            >>> results = await client.run_job(config)
+            >>> print(results.best_script.preflight_script)
+        """
+        job_id = await self.create_job(config)
+        
+        await self.wait_for_completion(
+            job_id,
+            poll_interval=poll_interval,
+            timeout=timeout,
+            on_event=on_event,
+        )
+        
+        return await self.get_results(job_id)
+
+
+# -----------------------------------------------------------------------------
+# Synchronous Wrappers
+# -----------------------------------------------------------------------------
+
+def create_job(
+    config: ContextLearningJobConfig | Dict[str, Any],
+    base_url: str,
+    api_key: str,
+) -> str:
+    """Synchronous wrapper to create a context learning job.
+    
+    Args:
+        config: Job configuration
+        base_url: Backend API base URL
+        api_key: API key for authentication
+        
+    Returns:
+        Job ID
+    """
+    client = ContextLearningClient(base_url, api_key)
+    return asyncio.run(client.create_job(config))
+
+
+def get_job_status(
+    job_id: str,
+    base_url: str,
+    api_key: str,
+) -> ContextLearningJobStatus:
+    """Synchronous wrapper to get job status.
+    
+    Args:
+        job_id: Job ID
+        base_url: Backend API base URL
+        api_key: API key for authentication
+        
+    Returns:
+        Job status
+    """
+    client = ContextLearningClient(base_url, api_key)
+    return asyncio.run(client.get_job(job_id))
+
+
+def get_best_script(
+    job_id: str,
+    base_url: str,
+    api_key: str,
+) -> BestScriptResult:
+    """Synchronous wrapper to get best script.
+    
+    Args:
+        job_id: Job ID
+        base_url: Backend API base URL
+        api_key: API key for authentication
+        
+    Returns:
+        Best script result
+    """
+    client = ContextLearningClient(base_url, api_key)
+    return asyncio.run(client.get_best_script(job_id))
+
+
+def run_job(
+    config: ContextLearningJobConfig | Dict[str, Any],
+    base_url: str,
+    api_key: str,
+    *,
+    poll_interval: float = 5.0,
+    timeout: Optional[float] = None,
+    on_event: Optional[callable] = None,
+) -> ContextLearningResults:
+    """Synchronous wrapper to run a complete job.
+    
+    Args:
+        config: Job configuration
+        base_url: Backend API base URL
+        api_key: API key for authentication
+        poll_interval: Seconds between status checks
+        timeout: Maximum seconds to wait
+        on_event: Optional callback for each event
+        
+    Returns:
+        Complete results
+    """
+    client = ContextLearningClient(base_url, api_key)
+    return asyncio.run(
+        client.run_job(
+            config,
+            poll_interval=poll_interval,
+            timeout=timeout,
+            on_event=on_event,
+        )
+    )
+