DSPyMator
centimators.model_estimators.dspymator
DSPyMator: A scikit-learn compatible wrapper for DSPy modules.
DSPyMator
dataclass
Bases: TransformerMixin, BaseEstimator
DSPyMator is a scikit-learn compatible wrapper for DSPy modules.
Integrates DSPy programs (e.g., Predict, ChainOfThought) into the
centimators ecosystem so LLM predictions work with sklearn pipelines,
cross-validation, hyperparameter search, calibration and stacking. The
estimator is dataframe-agnostic through narwhals, accepting Polars, Pandas,
or numpy arrays. Input columns map to the program's input fields by
feature_names, column names, or position.
Decision outputs
When the target is a DSPy decision output, DSPyMator behaves like a
native sklearn classifier or regressor, whichever backend answers it
(Jev via dspy.LM("typesafe:jev-latest"), a hosted LLM, or a local
model served through dspy.LM):
Noulorbooltargets: binary classifier,classes_ = [False, True].ChoiceorLiteral[...]targets: multiclass classifier over the declared options.Scoretargets: regressor on the continuous expected level.
Classifiers expose predict_proba and classes_, so log_loss,
roc_auc, CalibratedClassifierCV and stacking work unchanged.
Probabilities are the backend's raw distribution, before thresholds or
weights. On Jev they are measured; on generative LMs the model reports
them.
Optimization
Pass any DSPy optimizer instance as optimizer (e.g., dspy.GEPA,
dspy.BootstrapFewShot, dspy.experimental.ReAnchor). fit compiles
a copy of the program with a copy of the optimizer and stores the
result in program_, so the parameters are never modified and the
estimator clones cleanly for GridSearchCV and cross_val_score.
Some optimizers (e.g., GEPA, MIPROv2) want a validation set. Pass
validation_data to fit as (X_val, y_val) or as a fraction of
the training data to hold out.
Execution
Predictions run asynchronously with bounded concurrency by default
(use_async=True). Inside a running event loop (e.g., Jupyter), async
mode needs nest_asyncio; otherwise it falls back to sync.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
program
|
Module
|
DSPy module with a signature defining input and output fields. |
required |
target_names
|
str | list[str]
|
Output field name(s) returned by |
required |
feature_names
|
list[str] | None
|
Column names mapped, in order, to the signature's input fields. If None, dataframe columns are used; numpy columns map to the input fields by position. |
None
|
lm
|
str | LM | None
|
A |
None
|
temperature
|
float | None
|
Sampling temperature, used only when |
None
|
max_tokens
|
int | None
|
Maximum response tokens, used only when |
None
|
optimizer
|
Any
|
Optional DSPy optimizer instance run during |
None
|
use_async
|
bool
|
Whether to predict asynchronously. Defaults to True. |
True
|
max_concurrent
|
int
|
Maximum concurrent requests when |
50
|
verbose
|
bool
|
Whether to show progress bars (requires |
True
|
Examples:
import dspy
from typing import Literal
from sklearn.model_selection import cross_val_score
from centimators.model_estimators import DSPyMator
class Sentiment(dspy.Signature):
"""Classify the sentiment of a financial headline."""
headline: str = dspy.InputField()
sentiment: Literal["bearish", "neutral", "bullish"] = dspy.OutputField(
desc="Market sentiment of the headline."
)
clf = DSPyMator(
program=dspy.Predict(Sentiment),
target_names="sentiment",
lm="typesafe:jev-latest",
)
cross_val_score(clf, X, y, cv=5, scoring="neg_log_loss")
Source code in src/centimators/model_estimators/dspymator.py
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 | |
fit(X, y=None, validation_data=None)
Fit the estimator, running optimizer on a copy of program if set.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
X
|
Training data (dataframe or numpy array). |
required | |
y
|
Target values. Optional when no optimizer is set. |
None
|
|
validation_data
|
tuple[Any, Any] | float | None
|
Validation data for optimizers that use it, as
|
None
|
Returns:
| Name | Type | Description |
|---|---|---|
self |
The fitted estimator. |
Source code in src/centimators/model_estimators/dspymator.py
predict(X)
Target predictions as a numpy array.
One target returns a 1-D array (bool or option labels for classifiers,
floats for Score regressors); several targets return a 2-D array.
Source code in src/centimators/model_estimators/dspymator.py
predict_proba(X)
Class probabilities, columns ordered as classes_.
Available when the single target is a Noul/bool or
Choice/Literal decision output.
Source code in src/centimators/model_estimators/dspymator.py
score(X, y, sample_weight=None)
Accuracy for classifiers, R^2 for Score regressors.
Source code in src/centimators/model_estimators/dspymator.py
transform(X, y=None)
All output fields of the program as a dataframe (decisions as values).