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

synth_ai/cli/local/experiment_queue/results.py

@@ -0,0 +1,373 @@
+"""Utilities for parsing job artifacts into structured summaries."""
+
+from __future__ import annotations
+
+import json
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, Iterable
+
+
+def _load_latest_json(directory: Path, pattern: str) -> tuple[dict[str, Any], Path] | tuple[None, None]:
+    candidates = sorted(directory.glob(pattern), key=lambda p: p.stat().st_mtime, reverse=True)
+    for candidate in candidates:
+        try:
+            with open(candidate, encoding="utf-8") as fh:
+                return json.load(fh), candidate
+        except Exception:
+            continue
+    return None, None
+
+
+@dataclass(slots=True)
+class LearningCurvePoint:
+    rollout_count: int | None = None
+    performance: float | None = None
+    checkpoint_pct: float | None = None
+    metadata: dict[str, Any] = field(default_factory=dict)
+
+    def to_dict(self) -> dict[str, Any]:
+        return {
+            "rollout_count": self.rollout_count,
+            "performance": self.performance,
+            "checkpoint_pct": self.checkpoint_pct,
+            "metadata": self.metadata,
+        }
+
+
+@dataclass(slots=True)
+class ResultSummary:
+    """Structured representation of job outputs used for DB + CLI."""
+
+    stdout: str = ""
+    stderr: str = ""
+    returncode: int = 0
+    stats: dict[str, Any] = field(default_factory=dict)
+    learning_curve_points: list[LearningCurvePoint] = field(default_factory=list)
+    best_score: float | None = None
+    baseline_score: float | None = None
+    total_rollouts: int | None = None
+    total_time: float | None = None
+    artifacts: dict[str, Any] = field(default_factory=dict)
+
+    def to_dict(self) -> dict[str, Any]:
+        return {
+            "returncode": self.returncode,
+            "stdout": self.stdout,
+            "stderr": self.stderr,
+            "stats": self.stats,
+            "learning_curve": [point.to_dict() for point in self.learning_curve_points],
+            "best_score": self.best_score,
+            "baseline_score": self.baseline_score,
+            "total_rollouts": self.total_rollouts,
+            "total_time": self.total_time,
+            "artifacts": self.artifacts,
+        }
+
+
+def _parse_learning_curve(data: dict[str, Any]) -> list[LearningCurvePoint]:
+    curve = data.get("curve") or data.get("checkpoints") or []
+    points: list[LearningCurvePoint] = []
+    if isinstance(curve, Iterable):
+        for entry in curve:
+            if not isinstance(entry, dict):
+                continue
+            points.append(
+                LearningCurvePoint(
+                    rollout_count=entry.get("rollout_count") or entry.get("rollouts"),
+                    performance=entry.get("performance")
+                    or entry.get("score")
+                    or entry.get("aggregate_score"),
+                    checkpoint_pct=entry.get("checkpoint_pct"),
+                    metadata={k: v for k, v in entry.items() if k not in {"rollout_count", "performance", "checkpoint_pct"}},
+                )
+            )
+    return points
+
+
+def _extract_rollouts_from_output(stdout: str, stderr: str, results_folder: Path | None = None) -> int | None:
+    """Extract rollout count from stdout/stderr and log files by looking for rollout patterns."""
+    import re
+    
+    # Look for patterns like "[BANKING77_ROLLOUT] ... index=75" or "rollout.*index=(\d+)"
+    patterns = [
+        r'\[.*ROLLOUT\].*index=(\d+)',
+        r'rollout.*index=(\d+)',
+        r'rollout\s+(\d+)',
+        r'completed\s+(\d+)\s+rollouts?',
+        r'total.*rollouts?[:\s]+(\d+)',
+        r'"count":\s*(\d+).*rollout',  # JSON stats: "count": 59 in rollout_duration_stats
+        r'rollout.*"count":\s*(\d+)',  # JSON stats in rollout context
+    ]
+    
+    max_rollout = None
+    texts_to_check = [stdout, stderr]
+    
+    # Also check log files if results_folder is provided
+    if results_folder and results_folder.exists():
+        log_files = sorted(
+            results_folder.glob("*_log_*.log"),
+            key=lambda p: p.stat().st_mtime,
+            reverse=True,
+        )
+        import logging
+        logger = logging.getLogger(__name__)
+        logger.debug(f"Found {len(log_files)} log files in {results_folder}")
+        for log_file in log_files[:2]:  # Check up to 2 most recent log files
+            try:
+                log_content = log_file.read_text(encoding="utf-8", errors="ignore")
+                texts_to_check.append(log_content)
+                logger.debug(f"Read log file: {log_file.name} ({len(log_content)} chars)")
+            except Exception as e:
+                logger.warning(f"Failed to read log file {log_file}: {e}")
+    
+    for text in texts_to_check:
+        for pattern in patterns:
+            matches = re.findall(pattern, text, re.IGNORECASE)
+            if matches:
+                rollout_nums = [int(m) for m in matches if m.isdigit()]
+                if rollout_nums:
+                    max_rollout = max(max_rollout or 0, max(rollout_nums))
+    
+    # Also try to extract from JSON stats if present
+    if max_rollout is None:
+        for text in texts_to_check:
+            # Look for "rollout_duration_stats" with JSON containing "count"
+            # Format in file: rollout_duration_stats": "{\n  \"count\": 30,\n  ...}"
+            # When read, escaped quotes stay as \", so we need to match \"count\"
+            # Try pattern with escaped quotes first (most common case)
+            stats_match = re.search(
+                r'rollout_duration_stats.*?\\"count\\"\s*:\s*(\d+)',
+                text,
+                re.IGNORECASE | re.DOTALL,
+            )
+            if stats_match:
+                count = int(stats_match.group(1))
+                max_rollout = max(max_rollout or 0, count) if max_rollout else count
+            else:
+                # Fallback: Try unescaped quotes (in case file was processed)
+                stats_match2 = re.search(
+                    r'rollout_duration_stats.*?"count"\s*:\s*(\d+)',
+                    text,
+                    re.IGNORECASE | re.DOTALL,
+                )
+                if stats_match2:
+                    count = int(stats_match2.group(1))
+                    max_rollout = max(max_rollout or 0, count) if max_rollout else count
+                else:
+                    # Last resort: Get all \"count\": N or "count": N values and take the max reasonable one
+                    all_counts_escaped = re.findall(r'\\"count\\"\s*:\s*(\d+)', text, re.IGNORECASE)
+                    all_counts_normal = re.findall(r'"count"\s*:\s*(\d+)', text, re.IGNORECASE)
+                    all_counts = all_counts_escaped + all_counts_normal
+                    if all_counts:
+                        counts = [int(c) for c in all_counts if c.isdigit()]
+                        if counts:
+                            count = max(counts)
+                            # Only use if it's a reasonable rollout count (not 0 or 1)
+                            if count > 1:
+                                max_rollout = max(max_rollout or 0, count) if max_rollout else count
+    
+    return max_rollout if max_rollout is not None else None
+
+
+def _parse_text_results_file(results_file: Path) -> dict[str, Any]:
+    """Parse GEPA/MIPRO text result files to extract scores and metadata."""
+    import re
+    
+    result = {}
+    try:
+        content = results_file.read_text(encoding="utf-8")
+        
+        # Extract baseline score
+        baseline_match = re.search(r'Baseline Score:\s*([\d.]+)', content, re.IGNORECASE)
+        if baseline_match:
+            result["baseline_score"] = float(baseline_match.group(1))
+        
+        # Extract best score
+        best_match = re.search(r'Best Score:\s*([\d.]+)', content, re.IGNORECASE)
+        if best_match:
+            result["best_score"] = float(best_match.group(1))
+        
+        # Extract job ID
+        job_id_match = re.search(r'Job ID:\s*(\S+)', content, re.IGNORECASE)
+        if job_id_match:
+            result["job_id"] = job_id_match.group(1)
+            
+    except Exception:
+        pass
+    
+    return result
+
+
+def collect_result_summary(results_folder: Path, stdout: str = "", stderr: str = "") -> ResultSummary:
+    """Introspect result artifacts saved by prompt learning jobs.
+    
+    Args:
+        results_folder: Path to results directory
+        stdout: Standard output from job execution
+        stderr: Standard error from job execution
+        
+    Returns:
+        ResultSummary with parsed results
+        
+    Raises:
+        AssertionError: If inputs are invalid
+    """
+    from .validation import validate_path
+    
+    # Validate inputs
+    assert results_folder is not None, "results_folder cannot be None"
+    path = validate_path(results_folder, "results_folder", must_exist=False)
+    assert isinstance(stdout, str), (
+        f"stdout must be str, got {type(stdout).__name__}"
+    )
+    assert isinstance(stderr, str), (
+        f"stderr must be str, got {type(stderr).__name__}"
+    )
+    
+    summary = ResultSummary()
+    summary.stdout = stdout
+    summary.stderr = stderr
+    
+    if not path.exists():
+        # Try to extract rollouts from stdout/stderr even if no results folder
+        extracted = _extract_rollouts_from_output(stdout, stderr, results_folder=None)
+        if extracted is not None:
+            assert extracted > 0, f"total_rollouts must be > 0, got {extracted}"
+        summary.total_rollouts = extracted
+        return summary
+
+    learning_curve_data, curve_path = _load_latest_json(path, "*learning_curve*.json")
+    if learning_curve_data:
+        assert isinstance(learning_curve_data, dict), (
+            f"learning_curve_data must be dict, got {type(learning_curve_data).__name__}"
+        )
+        summary.learning_curve_points = _parse_learning_curve(learning_curve_data)
+        assert isinstance(summary.learning_curve_points, list), (
+            f"learning_curve_points must be list, got {type(summary.learning_curve_points).__name__}"
+        )
+        
+        points = [p for p in summary.learning_curve_points if p.performance is not None]
+        if points:
+            baseline = points[0].performance
+            assert baseline is not None and 0 <= baseline <= 1, (
+                f"baseline_score must be in [0, 1], got {baseline}"
+            )
+            summary.baseline_score = baseline
+            
+            best = max((p.performance for p in points if p.performance is not None), default=None)
+            if best is not None:
+                assert 0 <= best <= 1, f"best_score must be in [0, 1], got {best}"
+            summary.best_score = best
+            
+            if points[-1].rollout_count is not None:
+                assert points[-1].rollout_count > 0, (
+                    f"rollout_count must be > 0, got {points[-1].rollout_count}"
+                )
+                summary.total_rollouts = points[-1].rollout_count
+            elif learning_curve_data.get("total_rollouts") is not None:
+                total = learning_curve_data.get("total_rollouts")
+                assert isinstance(total, int) and total > 0, (
+                    f"total_rollouts must be int > 0, got {total}"
+                )
+                summary.total_rollouts = total
+        if curve_path:
+            assert isinstance(curve_path, Path), (
+                f"curve_path must be Path, got {type(curve_path).__name__}"
+            )
+            summary.artifacts["learning_curve_path"] = str(curve_path)
+
+    stats_data, stats_path = _load_latest_json(path, "*stats*.json")
+    if stats_data:
+        assert isinstance(stats_data, dict), (
+            f"stats_data must be dict, got {type(stats_data).__name__}"
+        )
+        summary.stats = stats_data
+        total_time = stats_data.get("total_time", summary.total_time)
+        if total_time is not None:
+            assert isinstance(total_time, int | float) and total_time >= 0, (
+                f"total_time must be >= 0, got {total_time}"
+            )
+        summary.total_time = total_time
+        
+        total_rollouts_from_stats = stats_data.get("total_rollouts")
+        if total_rollouts_from_stats is not None:
+            assert isinstance(total_rollouts_from_stats, int) and total_rollouts_from_stats > 0, (
+                f"total_rollouts must be int > 0, got {total_rollouts_from_stats}"
+            )
+        summary.total_rollouts = summary.total_rollouts or total_rollouts_from_stats
+        
+        if stats_path:
+            assert isinstance(stats_path, Path), (
+                f"stats_path must be Path, got {type(stats_path).__name__}"
+            )
+            summary.artifacts["stats_path"] = str(stats_path)
+
+    # If no JSON files found, try parsing text result files or extracting from stdout/stderr/log files
+    if summary.total_rollouts is None:
+        import logging
+        logger = logging.getLogger(__name__)
+        logger.info(
+            "No rollouts found in JSON files, attempting extraction from stdout/stderr/log files. "
+            "Results folder: %s",
+            path,
+        )
+        extracted_rollouts = _extract_rollouts_from_output(stdout, stderr, results_folder=path)
+        if extracted_rollouts is not None:
+            assert extracted_rollouts > 0, (
+                f"extracted_rollouts must be > 0, got {extracted_rollouts}"
+            )
+            summary.total_rollouts = extracted_rollouts
+            logger.info(
+                "✅ Extracted total_rollouts=%d from stdout/stderr/log files",
+                extracted_rollouts,
+            )
+        else:
+            logger.warning(
+                "❌ Failed to extract rollouts from stdout/stderr/log files. "
+                "Stdout length: %d, Stderr length: %d, Results folder exists: %s",
+                len(stdout),
+                len(stderr),
+                path.exists() if path else False,
+            )
+    
+    if summary.best_score is None or summary.baseline_score is None:
+        # Look for text result files (gepa_results_*.txt, mipro_results_*.txt)
+        result_files = sorted(
+            path.glob("*_results_*.txt"),
+            key=lambda p: p.stat().st_mtime,
+            reverse=True,
+        )
+        if result_files:
+            text_results = _parse_text_results_file(result_files[0])
+            assert isinstance(text_results, dict), (
+                f"_parse_text_results_file must return dict, got {type(text_results).__name__}"
+            )
+            
+            best_score = text_results.get("best_score")
+            if best_score is not None:
+                assert isinstance(best_score, int | float) and 0 <= best_score <= 1, (
+                    f"best_score must be in [0, 1], got {best_score}"
+                )
+                summary.best_score = float(best_score)
+            
+            baseline_score = text_results.get("baseline_score")
+            if baseline_score is not None:
+                assert isinstance(baseline_score, int | float) and 0 <= baseline_score <= 1, (
+                    f"baseline_score must be in [0, 1], got {baseline_score}"
+                )
+                summary.baseline_score = float(baseline_score)
+            
+            if result_files[0]:
+                summary.artifacts["results_txt_path"] = str(result_files[0])
+
+    best_prompt_candidates = sorted(
+        path.glob("*best_prompt*.json"),
+        key=lambda p: p.stat().st_mtime,
+        reverse=True,
+    )
+    best_prompt_file = next(iter(best_prompt_candidates), None)
+    if best_prompt_file:
+        summary.artifacts["best_prompt_path"] = str(best_prompt_file)
+
+    return summary

synth_ai/cli/local/experiment_queue/schemas.py

@@ -0,0 +1,131 @@
+"""Pydantic schemas shared between CLI, worker, and API surfaces."""
+
+from __future__ import annotations
+
+from datetime import datetime
+from typing import Any
+
+from pydantic import AliasChoices, BaseModel, ConfigDict, Field, field_validator
+
+from .models import (
+    Experiment,
+    ExperimentJob,
+    ExperimentJobStatus,
+    ExperimentJobType,
+    ExperimentStatus,
+    Trial,
+    TrialStatus,
+)
+
+
+class ExperimentJobSpec(BaseModel):
+    """Job specification supplied when submitting an experiment."""
+
+    job_type: ExperimentJobType = Field(
+        default=ExperimentJobType.GEPA,
+        validation_alias=AliasChoices("job_type", "type"),
+        serialization_alias="job_type",
+    )
+    config_path: str
+    config_overrides: dict[str, Any] = Field(default_factory=dict)
+
+
+class ExperimentSubmitRequest(BaseModel):
+    """Submission payload for a new experiment."""
+
+    name: str
+    description: str | None = None
+    parallelism: int = Field(default=3, ge=1, le=8)
+    jobs: list[ExperimentJobSpec]
+    metadata: dict[str, Any] = Field(default_factory=dict)
+
+    @field_validator("jobs")
+    @classmethod
+    def _ensure_jobs(cls, value: list[ExperimentJobSpec]) -> list[ExperimentJobSpec]:
+        if not value:
+            raise ValueError("At least one job specification is required.")
+        return value
+
+
+class ExperimentJobSummary(BaseModel):
+    """Serializable summary of a job."""
+
+    job_id: str
+    job_type: ExperimentJobType
+    status: ExperimentJobStatus
+    created_at: datetime
+    started_at: datetime | None = None
+    completed_at: datetime | None = None
+    backend_job_id: str | None = None
+    error: str | None = None
+
+    model_config = ConfigDict(from_attributes=True, use_enum_values=True)
+
+    @classmethod
+    def from_job(cls, job: ExperimentJob) -> ExperimentJobSummary:
+        return cls(
+            job_id=job.job_id,
+            job_type=ExperimentJobType(job.job_type),
+            status=ExperimentJobStatus(job.status),
+            created_at=job.created_at,
+            started_at=job.started_at,
+            completed_at=job.completed_at,
+            backend_job_id=job.backend_job_id,
+            error=job.error,
+        )
+
+
+class TrialSummary(BaseModel):
+    """Serializable trial summary."""
+
+    trial_id: str
+    trial_number: int | None
+    system_name: str | None
+    reward: float | None
+    aggregate_score: float | None
+    status: TrialStatus
+
+    model_config = ConfigDict(from_attributes=True, use_enum_values=True)
+
+    @classmethod
+    def from_trial(cls, trial: Trial) -> TrialSummary:
+        return cls(
+            trial_id=trial.trial_id,
+            trial_number=trial.trial_number,
+            system_name=trial.system_name,
+            reward=trial.reward,
+            aggregate_score=trial.aggregate_score,
+            status=TrialStatus(trial.status),
+        )
+
+
+class ExperimentSummary(BaseModel):
+    """Serializable experiment summary for CLI dashboards."""
+
+    experiment_id: str
+    name: str
+    status: ExperimentStatus
+    description: str | None = None
+    parallelism_limit: int
+    created_at: datetime
+    started_at: datetime | None = None
+    completed_at: datetime | None = None
+    job_count: int = 0
+    trial_count: int = 0
+
+    model_config = ConfigDict(from_attributes=True, use_enum_values=True)
+
+    @classmethod
+    def from_experiment(cls, experiment: Experiment) -> ExperimentSummary:
+        return cls(
+            experiment_id=experiment.experiment_id,
+            name=experiment.name,
+            status=ExperimentStatus(experiment.status),
+            description=experiment.description,
+            parallelism_limit=experiment.parallelism_limit,
+            created_at=experiment.created_at,
+            started_at=experiment.started_at,
+            completed_at=experiment.completed_at,
+            job_count=len(experiment.jobs),
+            trial_count=len(experiment.trials),
+        )