Extend nyx-probing with automated variance collection using nyx-substrate for database persistence. Adds 4 new CLI commands for collecting and analyzing variance in Echo Probe measurements. New Features: - VarianceRunner: Automated 1000x probe collection with progress bars - 4 new CLI commands: - variance collect: Single term variance collection - variance batch: Batch collection from glossary files - variance stats: View session statistics - variance analyze: Cross-session variance analysis - Full integration with nyx-substrate database layer Files Added: - nyx_probing/runners/__init__.py: Runner module initialization - nyx_probing/runners/variance_runner.py: VarianceRunner class (~200 LOC) - nyx_probing/cli/variance.py: CLI commands (~250 LOC) Files Modified: - pyproject.toml: Added nyx-substrate>=0.1.0 dependency - nyx_probing/cli/probe.py: Registered variance command group - data/glossary/master.json: Updated from scanning Integration: - nyx-probing → nyx-substrate → phoebe (PostgreSQL) - Stores each probe run in variance_probe_runs table - Rich progress bars and statistics display - Session-based tracking with UUIDs Usage Examples: nyx-probe variance collect "Geworfenheit" --runs 1000 nyx-probe variance batch depth_3_champions.json nyx-probe variance stats <SESSION_ID> nyx-probe variance analyze "Geworfenheit" Status: Phase 1B complete, ready for baseline collection 🌙💜 Generated with Claude Code Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
230 lines
7.6 KiB
Python
230 lines
7.6 KiB
Python
"""
|
|
Variance collection runner for Echo Probe.
|
|
|
|
Automates running Echo Probe 1000x to measure variance in depth,
|
|
echo types, and chain patterns for baseline characterization.
|
|
"""
|
|
|
|
import uuid
|
|
from typing import List, Dict, Any
|
|
from rich.progress import Progress, SpinnerColumn, TextColumn, BarColumn, TaskProgressColumn
|
|
from rich.console import Console
|
|
|
|
from ..core.model import NyxModel
|
|
from ..probes.echo_probe import EchoProbe
|
|
from nyx_substrate.database import PhoebeConnection, VarianceProbeDAO
|
|
from nyx_substrate.schemas import VarianceProbeRun
|
|
|
|
|
|
console = Console()
|
|
|
|
|
|
class VarianceRunner:
|
|
"""
|
|
Automated variance collection for Echo Probe.
|
|
|
|
Runs Echo Probe N times on a term, storing each result in phoebe
|
|
for variance analysis.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
model: NyxModel,
|
|
dao: VarianceProbeDAO,
|
|
max_rounds: int = 3,
|
|
max_new_tokens: int = 50,
|
|
temperature: float = 0.8,
|
|
):
|
|
"""
|
|
Initialize VarianceRunner.
|
|
|
|
Args:
|
|
model: Loaded NyxModel
|
|
dao: VarianceProbeDAO for database storage
|
|
max_rounds: Max echo rounds per probe
|
|
max_new_tokens: Max tokens per generation
|
|
temperature: Sampling temperature
|
|
"""
|
|
self.model = model
|
|
self.dao = dao
|
|
self.probe = EchoProbe(
|
|
model=model,
|
|
max_rounds=max_rounds,
|
|
max_new_tokens=max_new_tokens,
|
|
temperature=temperature,
|
|
)
|
|
self.max_rounds = max_rounds
|
|
self.max_new_tokens = max_new_tokens
|
|
self.temperature = temperature
|
|
|
|
def run_session(
|
|
self,
|
|
term: str,
|
|
runs: int = 1000,
|
|
show_progress: bool = True,
|
|
) -> uuid.UUID:
|
|
"""
|
|
Run variance collection session on a single term.
|
|
|
|
Args:
|
|
term: Term to probe
|
|
runs: Number of runs (default: 1000)
|
|
show_progress: Show progress bar
|
|
|
|
Returns:
|
|
session_id UUID
|
|
|
|
Example:
|
|
>>> runner = VarianceRunner(model, dao)
|
|
>>> session_id = runner.run_session("Geworfenheit", runs=1000)
|
|
>>> print(f"Session: {session_id}")
|
|
"""
|
|
session_id = uuid.uuid4()
|
|
|
|
console.print(f"\n[bold cyan]🔬 Variance Collection Session[/bold cyan]")
|
|
console.print(f"Term: [bold]{term}[/bold]")
|
|
console.print(f"Runs: {runs}")
|
|
console.print(f"Session ID: {session_id}\n")
|
|
|
|
if show_progress:
|
|
with Progress(
|
|
SpinnerColumn(),
|
|
TextColumn("[progress.description]{task.description}"),
|
|
BarColumn(),
|
|
TaskProgressColumn(),
|
|
console=console,
|
|
) as progress:
|
|
task = progress.add_task(f"Probing '{term}'...", total=runs)
|
|
|
|
for run_number in range(1, runs + 1):
|
|
# Run probe
|
|
result = self.probe.probe(term)
|
|
|
|
# Convert echo types to strings
|
|
echo_types_str = [et.name for et in result.echo_types]
|
|
|
|
# Store in phoebe
|
|
self.dao.insert_run(
|
|
session_id=session_id,
|
|
term=term,
|
|
run_number=run_number,
|
|
depth=result.depth,
|
|
rounds=result.rounds,
|
|
echo_types=echo_types_str,
|
|
chain=result.chain,
|
|
model_name=self.model.model_name,
|
|
temperature=self.temperature,
|
|
max_rounds=self.max_rounds,
|
|
max_new_tokens=self.max_new_tokens,
|
|
)
|
|
|
|
progress.update(task, advance=1)
|
|
else:
|
|
# No progress bar
|
|
for run_number in range(1, runs + 1):
|
|
result = self.probe.probe(term)
|
|
echo_types_str = [et.name for et in result.echo_types]
|
|
|
|
self.dao.insert_run(
|
|
session_id=session_id,
|
|
term=term,
|
|
run_number=run_number,
|
|
depth=result.depth,
|
|
rounds=result.rounds,
|
|
echo_types=echo_types_str,
|
|
chain=result.chain,
|
|
model_name=self.model.model_name,
|
|
temperature=self.temperature,
|
|
max_rounds=self.max_rounds,
|
|
max_new_tokens=self.max_new_tokens,
|
|
)
|
|
|
|
console.print(f"\n✅ [bold green]Session complete![/bold green]")
|
|
console.print(f"Stored {runs} runs in phoebe")
|
|
console.print(f"Session ID: [bold]{session_id}[/bold]\n")
|
|
|
|
return session_id
|
|
|
|
def run_batch(
|
|
self,
|
|
terms: List[str],
|
|
runs_per_term: int = 1000,
|
|
show_progress: bool = True,
|
|
) -> Dict[str, uuid.UUID]:
|
|
"""
|
|
Run variance collection on multiple terms.
|
|
|
|
Args:
|
|
terms: List of terms to probe
|
|
runs_per_term: Number of runs per term
|
|
show_progress: Show progress bar
|
|
|
|
Returns:
|
|
Dictionary mapping term -> session_id
|
|
|
|
Example:
|
|
>>> runner = VarianceRunner(model, dao)
|
|
>>> sessions = runner.run_batch(["Geworfenheit", "Vernunft"], runs_per_term=1000)
|
|
"""
|
|
console.print(f"\n[bold cyan]🔬 Batch Variance Collection[/bold cyan]")
|
|
console.print(f"Terms: {len(terms)}")
|
|
console.print(f"Runs per term: {runs_per_term}")
|
|
console.print(f"Total runs: {len(terms) * runs_per_term}\n")
|
|
|
|
sessions = {}
|
|
|
|
for idx, term in enumerate(terms, 1):
|
|
console.print(f"[bold]Term {idx}/{len(terms)}:[/bold] {term}")
|
|
session_id = self.run_session(term, runs=runs_per_term, show_progress=show_progress)
|
|
sessions[term] = session_id
|
|
|
|
console.print(f"\n✅ [bold green]Batch complete![/bold green]")
|
|
console.print(f"Collected variance for {len(terms)} terms")
|
|
|
|
return sessions
|
|
|
|
def get_session_summary(self, session_id: uuid.UUID) -> Dict[str, Any]:
|
|
"""
|
|
Get summary statistics for a session.
|
|
|
|
Args:
|
|
session_id: Session UUID
|
|
|
|
Returns:
|
|
Dictionary with statistics
|
|
|
|
Example:
|
|
>>> summary = runner.get_session_summary(session_id)
|
|
>>> print(f"Average depth: {summary['avg_depth']}")
|
|
"""
|
|
return self.dao.get_session_stats(session_id)
|
|
|
|
def display_session_stats(self, session_id: uuid.UUID) -> None:
|
|
"""
|
|
Display session statistics to console.
|
|
|
|
Args:
|
|
session_id: Session UUID
|
|
"""
|
|
stats = self.get_session_summary(session_id)
|
|
|
|
console.print(f"\n[bold cyan]📊 Session Statistics[/bold cyan]")
|
|
console.print(f"Session ID: {session_id}")
|
|
console.print(f"Term: [bold]{stats['term']}[/bold]")
|
|
console.print(f"Total runs: {stats['total_runs']}")
|
|
console.print(f"Average depth: {stats['avg_depth']:.2f}")
|
|
console.print(f"Average rounds: {stats['avg_rounds']:.2f}")
|
|
|
|
console.print("\n[bold]Depth Distribution:[/bold]")
|
|
dist = stats['depth_distribution']
|
|
for depth_val in ['0', '1', '2', '3']:
|
|
count = dist.get(depth_val, 0)
|
|
pct = (count / stats['total_runs'] * 100) if stats['total_runs'] > 0 else 0
|
|
console.print(f" Depth {depth_val}: {count:4d} ({pct:5.1f}%)")
|
|
|
|
console.print("\n[bold]Most Common Echo Types:[/bold]")
|
|
for echo_info in stats['most_common_echo_types'][:5]:
|
|
console.print(f" {echo_info['type']:12s}: {echo_info['count']:4d}")
|
|
|
|
console.print()
|