Spaces:
Sleeping
Sleeping
File size: 8,452 Bytes
fede53c | 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 | """Abstract base class for all ODSE tasks.
Every concrete task (cleaning, feature engineering, ...) inherits from
``BaseTask`` and implements the abstract hooks. The base class provides:
* ``setup()`` - initialises / resets the state.
* ``execute()`` =. validates + dispatches an action, computes reward.
* ``build_observation()`` - constructs the ``Observation`` pydantic model.
* ``calculate_reward()`` - sklearn proxy model accuracy (5-fold CV).
"""
from __future__ import annotations
from abc import ABC, abstractmethod
from typing import Any, Dict, List, Set
import numpy as np
import pandas as pd
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_val_score
from sklearn.preprocessing import LabelEncoder
from ..data.data_manager import DataState
from ..data.datasets import DatasetConfig
from ..models import (
Action,
ColumnInfo,
Difficulty,
Observation,
StepResult,
SubmitAction,
TaskType,
)
class BaseTask(ABC):
"""Base class that every ODSE task must inherit from.
Subclass contract
-----------------
* ``TASK_TYPE``: the ``TaskType`` enum member.
* ``SUPPORTED_ACTIONS``: set of action_types literal strings.
* ``apply_action()``: mutate ``self.data_state`` for a given action.
* ``calculate_reward()``: scaler reward for one step.
* ``is_done()``: whether the episode should terminate.
* ``grade()``: score the final state (dict with ``"score"``).
* ``get_goal_description()``: human readable goal string
"""
# -- subclasses must set these class-level attributes -------------------
TASK_TYPE: TaskType
SUPPORTED_ACTIONS: Set[str]
#Default per-difficulty step limits (overridable per task)
MAX_STEPS : Dict[Difficulty, int] = {
Difficulty.EASY: 15,
Difficulty.MEDIUM: 30,
Difficulty.HARD: 45
}
def __init__(
self,
dataset_config: DatasetConfig,
difficulty: Difficulty,
seed: int = 42,
) -> None:
self.dataset_config = dataset_config
self.difficulty = difficulty
self.seed = seed
self.max_steps: int = self.MAX_STEPS.get(difficulty, 25)
#State - populated on setup()
self.data_state: DataState | None = None
self.step_count: int = 0
self._initial_accuracy: float = 0.0
self._previous_accuracy: float = 0.0
# Lifecycle
def setup(self) -> Observation:
"""Initialize / reset the task. Called by ``ODSEnvironment.reset()``."""
self.data_state = DataState(self.dataset_config.df, name='initial')
self.step_count = 0
self._initial_accuracy = self.calculate_accuracy()
self._previous_accuracy = self._initial_accuracy
return self.build_observation()
# Action Dispatch
def execute(self, action: Action) -> StepResult:
"""Validate, apply, *action*, compute reward, and return ``StepResult``."""
assert self.data_state is not None, "Call setup() before executing actions"
self.step_count += 1
# ----- Submit terminates immediately ------
if isinstance(action, SubmitAction):
obs = self.build_observation()
return StepResult(
observation=obs,
reward=0.0,
done=True,
info={"reason": "Submit"}
)
# ----- Validate action type for this task -----
if action.action_type not in self.SUPPORTED_ACTIONS:
obs = self.build_observation()
return StepResult(
observation=obs,
reward=-0.1, # Penalty for invalid action
done=False,
info={
"error": (
f"Action `{action.action_type}` not supported "
f"by {self.TASK_TYPE.value}"
)
}
)
#can be optimized with prev acc
old_accuracy = self.calculate_accuracy()
# ---- Delegate to the concreate task ----------
self.apply_action(action)
new_accuracy = self.calculate_accuracy()
reward = self.calculate_reward(old_accuracy, new_accuracy, action)
self._previous_accuracy = new_accuracy
obs = self.build_observation()
done = self.is_done()
return StepResult(
observation=obs,
reward=reward,
done=done,
info={
"old_accuracy": old_accuracy,
"new_accuracy": new_accuracy,
"accuracy_delta": new_accuracy - old_accuracy
}
)
# Abstract interface - subclasses must implement these methods
@abstractmethod
def apply_action(self, action: Action) -> None:
"""Mutate ``self.data_state`` according to *action*"""
pass
@abstractmethod
def calculate_reward(self, old_accuracy: float, new_accuracy: float, action: Action) -> float:
"""Return the scalar reward for a single step"""
pass
@abstractmethod
def is_done(self) -> bool:
"""REturn ``True`` when the epiisode should end (besides submit)."""
pass
@abstractmethod
def grade(self) -> Dict[str, Any]:
"""Score the final state. Must return dict containiing ``"score"``."""
pass
@abstractmethod
def get_goal_description(self) -> str:
"""Return a human-readable string describing the agent's goal."""
pass
# Observation Builder
def build_observation(self) -> Observation:
"""Construct an ``Observation`` from the current ``DataState``."""
ds = self.data_state
df = ds.df
columns: List[ColumnInfo] = []
total_nulls = 0
for col in df.columns:
nc = int(df[col].isnull().sum())
total_nulls += nc
columns.append(
ColumnInfo(
name=col,
dtype=str(df[col].dtype),
null_count=nc,
null_percentage=(
round(nc / len(df) * 100, 2) if len(df) > 0 else 0.0
),
unique_count=int(df[col].nunique()),
is_numeric=pd.api.types.is_numeric_dtype(df[col]),
)
)
return Observation(
columns=columns,
sample_head=df.head().to_dict(orient="list"),
shape=tuple(df.shape),
current_accuracy=self.calculate_accuracy(),
step_count=self.step_count,
nulls_remaining=total_nulls,
task_type=self.TASK_TYPE.value,
difficulty=self.difficulty.value,
goal_description=self.get_goal_description(),
available_actions=sorted(self.SUPPORTED_ACTIONS | {"submit"})
)
# =====================================================================
# Proxy model accuracy (shared by all tasks)
# =====================================================================
def calculate_accuracy(self) -> float:
"""5-fold CV accuracy using a LogisticRegression proxy model.
Only complete (non-null) rows are used. Returns 0.0 when the
dataset is too small or the model fails.
"""
df = self.data_state.df.dropna()
target = self.dataset_config.target_column
features = [
c for c in self.dataset_config.feature_columns if c in df.columns
]
if len(df) < 10 or target not in df.columns or not features:
return 0.0
try:
X = df[features].copy()
y = df[target]
# Encode categorical / string columns
for col in X.columns:
if not pd.api.types.is_numeric_dtype(X[col]):
le = LabelEncoder()
X[col] = le.fit_transform(X[col].astype(str))
X_arr = X.values.astype(float)
n_folds = min(5, len(df))
model = LogisticRegression(
max_iter=1000,
random_state=self.seed,
solver="lbfgs",
)
scores = cross_val_score(
model, X_arr, y, cv=n_folds, scoring="accuracy",
)
return float(np.mean(scores))
except Exception:
return 0.0 |