File size: 8,138 Bytes
6835659 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 | """
Experiment Configuration
Defines configuration for controlled experiments with statistical rigor.
"""
from __future__ import annotations
from dataclasses import dataclass, field
from enum import Enum
from typing import List, Dict, Any, Optional
from pathlib import Path
import json
class PlannerMode(str, Enum):
"""Planning mode for generation."""
DIRECT = "direct" # Raw prompt → generators
SINGLE_PLANNER = "single_planner" # 1 LLM call → plan → generators
COUNCIL = "council" # 3 LLM calls → merge → generators
EXTENDED_PROMPT = "extended_prompt" # 1 LLM call with 3× token budget
class PerturbationType(str, Enum):
"""Perturbation conditions for sensitivity testing."""
BASELINE = "baseline" # Normal generation
WRONG_IMAGE = "wrong_image" # Image from different prompt
WRONG_AUDIO = "wrong_audio" # Audio from different prompt
SEMANTIC_SHIFT_25 = "semantic_shift_25" # 25% semantic mismatch
SEMANTIC_SHIFT_50 = "semantic_shift_50" # 50% semantic mismatch
SEMANTIC_SHIFT_75 = "semantic_shift_75" # 75% semantic mismatch
RANDOM_IMAGE = "random_image" # Completely random image
RANDOM_AUDIO = "random_audio" # Completely random audio
@dataclass
class ExperimentConfig:
"""
Configuration for controlled experiments.
Supports:
- Multiple planning modes (RQ2: planning effect)
- Multiple perturbation conditions (RQ1: MSCI sensitivity)
- Statistical parameters for hypothesis testing
"""
# Experiment identity
name: str = "default_experiment"
description: str = ""
# Sample sizes
n_prompts: int = 50 # Minimum for statistical power
n_seeds: int = 3 # Replications per prompt
# Experimental conditions
modes: List[PlannerMode] = field(
default_factory=lambda: [PlannerMode.DIRECT, PlannerMode.SINGLE_PLANNER]
)
perturbations: List[PerturbationType] = field(
default_factory=lambda: [
PerturbationType.BASELINE,
PerturbationType.WRONG_IMAGE,
PerturbationType.WRONG_AUDIO,
]
)
# Statistical parameters
alpha: float = 0.05 # Significance level
power: float = 0.80 # Target statistical power
min_effect_size: float = 0.5 # Minimum Cohen's d to detect
# Execution parameters
deterministic: bool = True
base_seed: int = 42
use_ollama: bool = True
output_dir: str = "runs/experiments"
# Resource tracking
track_tokens: bool = True
track_time: bool = True
@property
def total_runs(self) -> int:
"""Total number of experimental runs."""
return self.n_prompts * self.n_seeds * len(self.modes) * len(self.perturbations)
@property
def conditions(self) -> List[str]:
"""List of all condition keys (mode_perturbation)."""
return [
f"{mode.value}_{pert.value}"
for mode in self.modes
for pert in self.perturbations
]
def required_sample_size(self, effect_size: Optional[float] = None) -> int:
"""
Compute required sample size for given effect size using power analysis.
For paired t-test with alpha=0.05, power=0.80:
- d=0.5 (medium): N≈34
- d=0.8 (large): N≈15
- d=0.3 (small): N≈90
Uses approximation: N ≈ 2 * ((z_alpha + z_beta) / d)^2
"""
from scipy import stats
d = effect_size or self.min_effect_size
z_alpha = stats.norm.ppf(1 - self.alpha / 2)
z_beta = stats.norm.ppf(self.power)
n = 2 * ((z_alpha + z_beta) / d) ** 2
return int(n) + 1 # Round up
def validate(self) -> List[str]:
"""Validate configuration and return any warnings."""
warnings = []
required_n = self.required_sample_size()
if self.n_prompts < required_n:
warnings.append(
f"Sample size ({self.n_prompts}) may be underpowered. "
f"Recommended: {required_n} for effect size d={self.min_effect_size}"
)
if self.n_seeds < 2:
warnings.append(
"n_seeds < 2: No replication variance can be estimated"
)
if self.alpha > 0.10:
warnings.append(
f"High alpha ({self.alpha}): Increased false positive risk"
)
return warnings
def to_dict(self) -> Dict[str, Any]:
"""Convert to dictionary for serialization."""
return {
"name": self.name,
"description": self.description,
"n_prompts": self.n_prompts,
"n_seeds": self.n_seeds,
"modes": [m.value for m in self.modes],
"perturbations": [p.value for p in self.perturbations],
"alpha": self.alpha,
"power": self.power,
"min_effect_size": self.min_effect_size,
"deterministic": self.deterministic,
"base_seed": self.base_seed,
"use_ollama": self.use_ollama,
"output_dir": self.output_dir,
"track_tokens": self.track_tokens,
"track_time": self.track_time,
"total_runs": self.total_runs,
"conditions": self.conditions,
}
@classmethod
def from_dict(cls, data: Dict[str, Any]) -> "ExperimentConfig":
"""Create from dictionary."""
# Convert string enums back
if "modes" in data:
data["modes"] = [PlannerMode(m) for m in data["modes"]]
if "perturbations" in data:
data["perturbations"] = [PerturbationType(p) for p in data["perturbations"]]
# Remove computed fields
data.pop("total_runs", None)
data.pop("conditions", None)
return cls(**data)
def save(self, path: Path):
"""Save configuration to JSON file."""
path = Path(path)
path.parent.mkdir(parents=True, exist_ok=True)
with path.open("w", encoding="utf-8") as f:
json.dump(self.to_dict(), f, indent=2)
@classmethod
def load(cls, path: Path) -> "ExperimentConfig":
"""Load configuration from JSON file."""
with Path(path).open("r", encoding="utf-8") as f:
return cls.from_dict(json.load(f))
# Preset configurations for common experiment types
PRESETS = {
"rq1_sensitivity": ExperimentConfig(
name="RQ1: MSCI Sensitivity",
description="Test if MSCI is sensitive to controlled semantic perturbations",
n_prompts=50,
n_seeds=3,
modes=[PlannerMode.SINGLE_PLANNER],
perturbations=[
PerturbationType.BASELINE,
PerturbationType.WRONG_IMAGE,
PerturbationType.WRONG_AUDIO,
PerturbationType.SEMANTIC_SHIFT_25,
PerturbationType.SEMANTIC_SHIFT_50,
PerturbationType.SEMANTIC_SHIFT_75,
],
),
"rq2_planning": ExperimentConfig(
name="RQ2: Planning Effect",
description="Test if structured planning improves cross-modal alignment",
n_prompts=50,
n_seeds=3,
modes=[
PlannerMode.DIRECT,
PlannerMode.SINGLE_PLANNER,
PlannerMode.COUNCIL,
PlannerMode.EXTENDED_PROMPT,
],
perturbations=[PerturbationType.BASELINE],
),
"full_ablation": ExperimentConfig(
name="Full Ablation Study",
description="Complete ablation across all modes and perturbations",
n_prompts=50,
n_seeds=3,
modes=[
PlannerMode.DIRECT,
PlannerMode.SINGLE_PLANNER,
PlannerMode.COUNCIL,
PlannerMode.EXTENDED_PROMPT,
],
perturbations=[
PerturbationType.BASELINE,
PerturbationType.WRONG_IMAGE,
PerturbationType.WRONG_AUDIO,
],
),
"quick_test": ExperimentConfig(
name="Quick Test",
description="Small-scale test run",
n_prompts=5,
n_seeds=1,
modes=[PlannerMode.DIRECT, PlannerMode.SINGLE_PLANNER],
perturbations=[PerturbationType.BASELINE, PerturbationType.WRONG_IMAGE],
),
}
|