File size: 25,408 Bytes
2b9a95b | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 | """
CausalSCM - Abstract base class for Structural Causal Models.
Each experiment implements its own SCM subclass that defines:
- Environment variable sampling
- Causal relationships
- Effect computation
Key principle: SCM writes effects into DroneSheet middleware.
This is the ONLY way SCM affects the game.
Enhanced features:
- Explicit causal variable and structure equation registration
- do-intervention support
- Automatic sampling with topological ordering
- Multi-format export (text, graph, DOT, JSON Schema)
"""
from abc import ABC, abstractmethod
from dataclasses import dataclass, field
from typing import Dict, Any, Optional, List, Tuple, Callable, Set
import random
import copy
# Handle both development (api.middleware) and Docker (middleware) imports
try:
from ...middleware.drone_state import EnvironmentEffects
from ...middleware.drone_sheet import DroneSheet
except ImportError:
from middleware.drone_state import EnvironmentEffects
from middleware.drone_sheet import DroneSheet
# ============================================================
# Core Data Structures for Explicit Causal Structure
# ============================================================
@dataclass
class CausalVariable:
"""
Represents a variable in the causal model.
Attributes:
name: Unique identifier for the variable
var_type: One of 'exogenous', 'latent', 'observed', 'derived'
parents: List of parent variable names
description: Human-readable description
domain: Optional (min, max) range for the variable
"""
name: str
var_type: str = 'observed' # 'exogenous', 'latent', 'observed', 'derived'
parents: List[str] = field(default_factory=list)
description: str = ''
domain: Optional[Tuple[float, float]] = None
@dataclass
class StructuralEquation:
"""
Represents a structural equation in the causal model.
The function takes a dict of parent values and returns the variable value.
Attributes:
target: Name of the variable this equation computes
function: Callable that takes parent values dict and returns computed value
description: Human-readable description of the equation
"""
target: str
function: Callable[[Dict[str, Any]], Any]
description: str = ''
@dataclass
class EnvironmentState:
"""
Complete environment state from SCM sampling.
Separates visible and latent (hidden) variables.
"""
# Variables visible to agent
visible: Dict[str, float] = field(default_factory=dict)
# Latent variables (hidden from agent)
latent: Dict[str, float] = field(default_factory=dict)
# Derived variables (computed from visible/latent)
derived: Dict[str, float] = field(default_factory=dict)
def get(self, key: str, default: float = 0.0) -> float:
"""Get any variable by name."""
if key in self.visible:
return self.visible[key]
if key in self.latent:
return self.latent[key]
if key in self.derived:
return self.derived[key]
return default
def all_variables(self) -> Dict[str, float]:
"""Get all variables (for logging)."""
result = {}
result.update(self.visible)
result.update(self.latent)
result.update(self.derived)
return result
@property
def weather_pattern(self) -> float:
"""Convenience accessor for common latent variable."""
return self.latent.get('weather_pattern', 0.5)
class CausalSCM(ABC):
"""
Abstract base class for Structural Causal Models.
Each SCM defines:
1. What environment variables exist
2. How they are sampled (causal structure)
3. How they affect the drone (effects)
Subclasses implement:
- sample_environment(): Generate environment state
- apply_effects(): Write effects to DroneSheet
Enhanced features:
- Explicit variable and equation registration via CausalVariable/StructuralEquation
- do-intervention support for causal reasoning
- Automatic sampling with topological ordering
- Multi-format export (text, graph, DOT, JSON Schema)
Usage:
scm = MyExperimentSCM(config)
env = scm.sample_environment()
scm.apply_effects(drone_sheet, env)
# DroneSheet now has effects applied
# do-intervention
intervened_scm = scm.do('weather_pattern', 0.9)
env = intervened_scm.sample()
"""
def __init__(self, config: Dict[str, Any]):
"""
Initialize SCM with experiment configuration.
Args:
config: Experiment configuration dict
"""
self.config = config
# Legacy storage (for backward compatibility)
self._variables: Dict[str, Dict[str, Any]] = {}
self._causal_edges: List[Tuple[str, str]] = []
# New enhanced storage
self._causal_variables: Dict[str, CausalVariable] = {}
self._equations: Dict[str, StructuralEquation] = {}
self._interventions: Dict[str, Any] = {}
self._topological_order: Optional[List[str]] = None
# Parse SCM config section
scm_config = config.get('scm', {})
self._exogenous_config: Dict[str, Dict[str, Any]] = scm_config.get(
'exogenous_variables', {}
)
self._scm_parameters: Dict[str, Any] = scm_config.get('parameters', {})
# ============================================================
# Variable and Equation Registration
# ============================================================
def register_variable(self, var: CausalVariable) -> None:
"""
Register a causal variable in the SCM.
Args:
var: CausalVariable to register
"""
self._causal_variables[var.name] = var
# Also update legacy storage for backward compatibility
self._variables[var.name] = {
'name': var.name,
'type': var.var_type,
'description': var.description,
'domain': var.domain,
'parents': var.parents,
}
# Add causal edges from parents
for parent in var.parents:
edge = (parent, var.name)
if edge not in self._causal_edges:
self._causal_edges.append(edge)
# Invalidate cached topological order
self._topological_order = None
def register_equation(self, eq: StructuralEquation) -> None:
"""
Register a structural equation in the SCM.
Args:
eq: StructuralEquation to register
"""
self._equations[eq.target] = eq
def add_variable(
self,
name: str,
var_type: str = 'observed',
description: str = '',
**kwargs
) -> None:
"""
Register a variable in the SCM (legacy method).
For new code, prefer register_variable() with CausalVariable.
Args:
name: Variable name
var_type: 'latent', 'observed', or 'derived'
description: Human-readable description
**kwargs: Additional variable properties (parents, domain)
"""
parents = kwargs.pop('parents', [])
domain = kwargs.pop('domain', None)
var = CausalVariable(
name=name,
var_type=var_type,
parents=parents,
description=description,
domain=domain,
)
self.register_variable(var)
def add_causal_edge(self, cause: str, effect: str) -> None:
"""
Add a causal edge to the graph.
Args:
cause: Cause variable name
effect: Effect variable name
"""
edge = (cause, effect)
if edge not in self._causal_edges:
self._causal_edges.append(edge)
self._topological_order = None
# ============================================================
# do-Intervention Support
# ============================================================
def do(self, var_name: str, value: Any) -> 'CausalSCM':
"""
Perform do-intervention: do(X = x).
Returns a new SCM instance with the intervention applied.
When sampling, the intervened variable will be fixed to the given value.
Args:
var_name: Variable name to intervene on
value: Value to set the variable to
Returns:
New CausalSCM instance with intervention applied
"""
# Create a deep copy to avoid modifying the original
intervened = copy.copy(self)
intervened._interventions = dict(self._interventions)
intervened._interventions[var_name] = value
return intervened
def clear_interventions(self) -> None:
"""Clear all interventions."""
self._interventions = {}
def get_interventions(self) -> Dict[str, Any]:
"""Get current interventions."""
return dict(self._interventions)
# ============================================================
# Topological Ordering
# ============================================================
def _compute_topological_order(self) -> List[str]:
"""
Compute topological ordering of variables using Kahn's algorithm.
Returns:
List of variable names in topological order
"""
if self._topological_order is not None:
return self._topological_order
# Build adjacency list and in-degree count
in_degree: Dict[str, int] = {}
children: Dict[str, List[str]] = {}
all_vars = set(self._causal_variables.keys())
for var_name in all_vars:
in_degree[var_name] = 0
children[var_name] = []
for cause, effect in self._causal_edges:
if cause in all_vars and effect in all_vars:
children[cause].append(effect)
in_degree[effect] += 1
# Find all nodes with no incoming edges
queue = [v for v in all_vars if in_degree[v] == 0]
result = []
while queue:
node = queue.pop(0)
result.append(node)
for child in children.get(node, []):
in_degree[child] -= 1
if in_degree[child] == 0:
queue.append(child)
# Check for cycles
if len(result) != len(all_vars):
missing = all_vars - set(result)
raise ValueError(f"Cycle detected in causal graph. Remaining nodes: {missing}")
self._topological_order = result
return result
# ============================================================
# Automatic Sampling
# ============================================================
def sample(self, exogenous_values: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
"""
Sample all variables according to the causal structure.
Executes structural equations in topological order.
Respects any active interventions.
Args:
exogenous_values: Optional dict of exogenous variable values.
If not provided, _sample_exogenous() is called.
Returns:
Dict mapping variable names to sampled values
"""
if not self._causal_variables:
# Fall back to sample_environment for backward compatibility
env = self.sample_environment()
return env.all_variables()
values: Dict[str, Any] = {}
exogenous = exogenous_values or {}
for var_name in self._compute_topological_order():
var = self._causal_variables[var_name]
# Check for intervention
if var_name in self._interventions:
values[var_name] = self._interventions[var_name]
continue
# Check for provided exogenous value
if var_name in exogenous:
values[var_name] = exogenous[var_name]
continue
# Sample exogenous variables
if var.var_type == 'exogenous' or not var.parents:
values[var_name] = self._sample_exogenous(var_name, var)
continue
# Execute structural equation
if var_name in self._equations:
eq = self._equations[var_name]
parent_values = {p: values.get(p) for p in var.parents}
values[var_name] = eq.function(parent_values)
else:
# No equation registered, use default sampling
values[var_name] = self._sample_exogenous(var_name, var)
return values
def _sample_exogenous(self, var_name: str, var: CausalVariable) -> Any:
"""
Sample an exogenous variable.
First checks config for distribution specification, then falls back
to domain-based uniform sampling.
Args:
var_name: Variable name
var: CausalVariable definition
Returns:
Sampled value
"""
# Check if config specifies distribution for this variable
if var_name in self._exogenous_config:
return self._sample_from_config(var_name, self._exogenous_config[var_name])
# Fall back to domain-based sampling
if var.domain is not None:
return random.uniform(var.domain[0], var.domain[1])
return random.random()
def _sample_from_config(self, var_name: str, var_config: Dict[str, Any]) -> Any:
"""
Sample a variable based on config specification.
Supported distributions:
- uniform: params {low, high}
- normal: params {mean, std}
- bernoulli: params {p}
- constant: params {value}
Args:
var_name: Variable name (for error messages)
var_config: Config dict with 'distribution' and 'params'
Returns:
Sampled value
"""
dist = var_config.get('distribution', 'uniform')
params = var_config.get('params', {})
if dist == 'uniform':
low = params.get('low', 0.0)
high = params.get('high', 1.0)
return random.uniform(low, high)
elif dist == 'normal':
mean = params.get('mean', 0.0)
std = params.get('std', 1.0)
return random.gauss(mean, std)
elif dist == 'bernoulli':
p = params.get('p', 0.5)
return 1.0 if random.random() < p else 0.0
elif dist == 'constant':
return params.get('value', 0.0)
else:
raise ValueError(f"Unknown distribution '{dist}' for variable '{var_name}'")
def get_parameter(self, name: str, default: Any = None) -> Any:
"""
Get a parameter from SCM config.
Checks scm.parameters first, then falls back to top-level config.
Args:
name: Parameter name
default: Default value if not found
Returns:
Parameter value
"""
if name in self._scm_parameters:
return self._scm_parameters[name]
return self.config.get(name, default)
# ============================================================
# Export Methods
# ============================================================
def describe(self) -> str:
"""
Return human-readable description of SCM.
Groups variables by type and shows causal structure.
"""
lines = [
f"SCM: {self.__class__.__name__}",
f"Variables: {len(self._causal_variables) or len(self._variables)}",
f"Causal edges: {len(self._causal_edges)}",
"",
]
# Group variables by type
by_type: Dict[str, List[CausalVariable]] = {
'exogenous': [],
'latent': [],
'observed': [],
'derived': [],
}
if self._causal_variables:
for var in self._causal_variables.values():
if var.var_type in by_type:
by_type[var.var_type].append(var)
else:
by_type['observed'].append(var)
for var_type, vars_list in by_type.items():
if vars_list:
lines.append(f"{var_type.title()} Variables:")
for var in vars_list:
domain_str = f" [{var.domain[0]}, {var.domain[1]}]" if var.domain else ""
lines.append(f" - {var.name}{domain_str}: {var.description}")
if var.parents:
lines.append(f" Parents: {', '.join(var.parents)}")
lines.append("")
else:
# Legacy format
lines.append("Variables:")
for name, info in self._variables.items():
var_type = info.get('type', 'unknown')
desc = info.get('description', '')
lines.append(f" - {name} ({var_type}): {desc}")
lines.append("")
lines.append("Causal Structure:")
for cause, effect in self._causal_edges:
lines.append(f" {cause} -> {effect}")
return "\n".join(lines)
def get_causal_graph(self) -> Dict[str, Any]:
"""
Return causal graph structure for visualization.
Returns:
Dict with 'nodes', 'edges', 'variables', and 'metadata'
"""
if self._causal_variables:
nodes = [
{
'name': var.name,
'type': var.var_type,
'description': var.description,
'domain': var.domain,
'parents': var.parents,
}
for var in self._causal_variables.values()
]
else:
nodes = [
{'name': name, **info}
for name, info in self._variables.items()
]
edges = [
{'from': cause, 'to': effect}
for cause, effect in self._causal_edges
]
return {
'nodes': nodes,
'edges': edges,
'metadata': {
'scm_name': self.__class__.__name__,
'num_variables': len(nodes),
'num_edges': len(edges),
}
}
def to_dot(self) -> str:
"""
Export causal graph in DOT/Graphviz format.
Returns:
DOT format string
"""
lines = [
f"digraph {self.__class__.__name__} {{",
" rankdir=TB;",
" node [shape=ellipse];",
"",
]
# Add nodes with styling based on type
if self._causal_variables:
latent_vars = []
observed_vars = []
derived_vars = []
for var in self._causal_variables.values():
if var.var_type in ('latent', 'exogenous'):
latent_vars.append(var)
elif var.var_type == 'derived':
derived_vars.append(var)
else:
observed_vars.append(var)
if latent_vars:
lines.append(" // Latent/Exogenous variables (dashed)")
for var in latent_vars:
label = f"{var.name}\\n({var.var_type})"
lines.append(f' {var.name} [label="{label}", style=dashed];')
lines.append("")
if observed_vars:
lines.append(" // Observed variables")
for var in observed_vars:
lines.append(f' {var.name} [label="{var.name}"];')
lines.append("")
if derived_vars:
lines.append(" // Derived variables (filled)")
for var in derived_vars:
lines.append(f' {var.name} [label="{var.name}", style=filled, fillcolor=lightgray];')
lines.append("")
else:
for name in self._variables.keys():
lines.append(f' {name};')
lines.append("")
# Add edges
lines.append(" // Edges")
for cause, effect in self._causal_edges:
lines.append(f" {cause} -> {effect};")
lines.append("}")
return "\n".join(lines)
def to_json_schema(self) -> Dict[str, Any]:
"""
Export SCM as JSON Schema format.
Returns:
Dict with full SCM specification
"""
variables = {}
if self._causal_variables:
for var in self._causal_variables.values():
var_spec = {
'type': var.var_type,
'exogenous': var.var_type == 'exogenous' or not var.parents,
'description': var.description,
'parents': var.parents,
}
if var.domain:
var_spec['domain'] = list(var.domain)
if var.name in self._equations:
var_spec['equation'] = self._equations[var.name].description
variables[var.name] = var_spec
else:
for name, info in self._variables.items():
variables[name] = {
'type': info.get('type', 'observed'),
'exogenous': not info.get('parents', []),
'description': info.get('description', ''),
'parents': info.get('parents', []),
}
if info.get('domain'):
variables[name]['domain'] = list(info['domain'])
return {
'name': self.__class__.__name__,
'variables': variables,
'edges': [
[cause, effect]
for cause, effect in self._causal_edges
],
}
# ============================================================
# Abstract Methods (must be implemented by subclasses)
# ============================================================
@abstractmethod
def sample_environment(self, equipment: dict = None) -> EnvironmentState:
"""
Sample environment state according to causal structure.
This method should:
1. Sample latent variables first
2. Sample visible variables conditioned on latent
3. Compute derived variables
Args:
equipment: Optional equipment choices from agent (e.g., flight_profile).
SCMs can use this to let agent influence environment variables.
Returns:
EnvironmentState with all variables
"""
pass
@abstractmethod
def apply_effects(
self,
sheet: DroneSheet,
env: EnvironmentState
) -> None:
"""
Apply environment effects to DroneSheet.
This is the ONLY way SCM affects the game!
Implementation should:
1. Compute effects from environment state
2. Call sheet.apply_environment_effects(effects)
Args:
sheet: DroneSheet middleware to write to
env: EnvironmentState from sample_environment()
"""
pass
class BaseSCM(CausalSCM):
"""
Base SCM implementation with minimal common functionality.
Provides:
- apply_effects() that delegates to abstract _compute_effects()
- Helper method for linear interpolation
Subclasses MUST implement:
- sample_environment(): Define how to sample environment state
- _compute_effects(): Define the specific causal mechanism
This base class is intentionally minimal. All experiment-specific
logic (latent variables, causal structure, etc.) belongs in subclasses.
"""
def __init__(self, config: Dict[str, Any]):
super().__init__(config)
def apply_effects(
self,
sheet: DroneSheet,
env: EnvironmentState
) -> None:
"""
Apply effects by delegating to _compute_effects().
This is the standard flow - subclasses implement _compute_effects().
"""
effects = self._compute_effects(sheet, env)
sheet.apply_environment_effects(effects)
@abstractmethod
def _compute_effects(
self,
sheet: DroneSheet,
env: EnvironmentState
) -> EnvironmentEffects:
"""
Compute environment effects for this specific causal mechanism.
This is where each experiment defines its unique causal structure.
Subclasses MUST implement this method.
Args:
sheet: DroneSheet to read drone state from (if needed)
env: EnvironmentState with visible/latent/derived variables
Returns:
EnvironmentEffects to be applied to DroneSheet
"""
pass
# ========== Helper methods for subclasses ==========
@staticmethod
def _interpolate(
t: float,
value_at_0: float,
value_at_1: float
) -> float:
"""
Linear interpolation helper.
Args:
t: Interpolation parameter (0 to 1)
value_at_0: Value when t=0
value_at_1: Value when t=1
Returns:
Interpolated value
"""
return value_at_0 * (1 - t) + value_at_1 * t
|