ml-wrappers

0.6.3 verified Fri May 01 auth: no python

Machine Learning Wrappers SDK for Python (v0.6.3) provides wrapper classes to unify model outputs for interpretability and fairness tools. Active development by Microsoft.

pip install ml-wrappers

Common errors

error ImportError: cannot import name 'DatasetWrapper' from 'ml_wrappers' ↓

cause DatasetWrapper was moved or renamed in older versions.

fix

Ensure ml-wrappers >=0.4.0; use 'from ml_wrappers import DatasetWrapper'.

error ValueError: The model parameter must be callable ↓

cause ModelWrapper expects a callable (e.g., model.predict), not a string or non-callable object.

fix

Pass the actual model object or a lambda wrapper: ModelWrapper(model=lambda x: model.predict(x), ...).

error AttributeError: module 'numpy' has no attribute 'object' ↓

cause numpy >=1.24 removed numpy.object alias; code uses np.object.

fix

Upgrade ml-wrappers to >=0.6.0 or use Python's built-in 'object'.

error OSError: [Errno 24] Too many open files ↓

cause Opening many model files without closing or using context managers.

fix

Use 'with ModelWrapper(...)' context manager or manually close resources.

Warnings

breaking In v0.6.0, numpy and pandas were updated to >2.0. Older versions of these libraries may cause compatibility issues. ↓

fix Upgrade numpy to >=1.24 and pandas to >=2.0.

breaking scikit-learn OneHotEncoder parameter changed from 'sparse' to 'sparse_output' in later scikit-learn versions. ml-wrappers v0.5.6+ handles this, but older versions may break. ↓

fix Update ml-wrappers to >=0.5.6 or manually adjust scikit-learn version.

deprecated TensorFlow 1.x support is deprecated. TensorFlow wrapper may fail with TF 2.x if protobuf is incompatible. ↓

fix Use TensorFlow 2.x with compatible protobuf (e.g., protobuf <4.0).

gotcha OpenAI wrapper in v0.5.x requires openai <1.0.0 for compatibility. v0.5.4 fixed this, but v0.5.3 and earlier break with openai >=1.0.0. ↓

fix Use openai<1.0.0 or upgrade ml-wrappers to >=0.5.4.

Imports

ModelWrapper

wrong

from ml_wrappers.model.wrapper import ModelWrapper

correct

from ml_wrappers import ModelWrapper

Direct import from ml_wrappers is preferred and stable.

DatasetWrapper
```
from ml_wrappers import DatasetWrapper
```

Quickstart

Wrap a ML model with ModelWrapper to unify predict/predict_proba outputs.

from ml_wrappers import ModelWrapper
import pandas as pd

model = ...  # your trained model
X_test = pd.DataFrame({'feature': [1, 2, 3]})
wrapped = ModelWrapper(model=model, model_task='classification')
# wrapped.predict(X_test)  # returns structured output