RoPWR: Robust Piecewise Regression

1.2.0 · active · verified Thu Apr 16

RoPWR (Robust Piecewise Regression) is a Python library that implements robust piecewise regression using convex optimization techniques. It provides a flexible framework to model non-linear relationships with segments, supporting various solvers via CVXPY. The current version is 1.2.0, with releases occurring every few months, primarily driven by feature additions, improvements, and dependency updates.

Common errors

```
TypeError: force_all_finite received a sparse matrix... Expected an array, got a sparse matrix.
```
cause This error occurs when an older version of `ropwr` (prior to 1.2.0) is used with `scikit-learn` version 1.6 or newer. `scikit-learn` deprecated `force_all_finite` in favor of `ensure_all_finite`.

fix Upgrade `ropwr` to `1.2.0` or higher: `pip install --upgrade ropwr`. Alternatively, if you cannot upgrade `ropwr`, downgrade `scikit-learn` to a version prior to 1.6: `pip install 'scikit-learn<1.6'`.
```
AttributeError: 'RoPWR' object has no attribute 'max_iter'
```
cause You are attempting to set the `max_iter` parameter when initializing or fitting a `RoPWR` model, but your `ropwr` library version is older than 1.0.0.

fix Upgrade `ropwr` to version 1.0.0 or higher: `pip install --upgrade ropwr`. If you cannot upgrade, remove the `max_iter` parameter from your `RoPWR` instantiation.
```
cvxpy.error.SolverError: Solver 'OSQP' failed. Try another solver or set verbose=True for more details.
```
cause The default or specified CVXPY solver (OSQP in this case) failed to converge or solve the optimization problem. This can be due to problem conditioning, solver limitations, or specific data characteristics.

fix Try explicitly specifying a different solver in the `RoPWR` constructor, for example, `RoPWR(..., solver='Clarabel')` or `RoPWR(..., solver='SCS')`. Ensure you have the necessary solvers installed (e.g., `pip install 'ropwr[all]'`). For detailed debugging, set `verbose=True` in the `RoPWR` constructor.
```
ImportError: cannot import name 'RoPWR' from 'ropwr'
```
cause This usually means the `RoPWR` class is not found in the `ropwr` package. This could be a typo in the import statement or an issue with the installation.

fix Verify that `ropwr` is correctly installed (`pip show ropwr`). Ensure the import statement is exactly `from ropwr import RoPWR`. If you've just installed it, restarting your interpreter or IDE might resolve module caching issues.

Warnings

gotcha Incompatibility with scikit-learn 1.6+ due to `force_all_finite` deprecation.
Fix: Upgrade ropwr to version 1.2.0 or higher. If unable to upgrade ropwr, downgrade scikit-learn to <1.6.
gotcha Older versions of RoPWR might experience OSQP solver convergence issues, especially with complex problems.
Fix: Upgrade ropwr to version 0.4.0 or higher. If using an older version, try specifying a different solver (e.g., `solver='Clarabel'` or `solver='SCS'`) or use `solver='auto'`.
gotcha The `max_iter` parameter for controlling solver iterations was introduced in v1.0.0. Attempting to use it with older versions will raise an AttributeError.
Fix: Upgrade ropwr to version 1.0.0 or higher to utilize the `max_iter` parameter.
gotcha Key features like enforcing continuous derivatives at split points and extrapolation methods were added in v1.0.0.
Fix: Ensure ropwr is version 1.0.0 or higher if your application requires these features. Refer to the documentation for how to enable continuous derivatives or specific extrapolation methods.

Install

pip install ropwr Install base package
pip install 'ropwr[all]' # Installs common CVXPY solvers like Clarabel, SCS, OSQP, ECOS Install with all optional solvers

Imports

RoPWR
```
from ropwr import RoPWR
```

Quickstart

This quickstart demonstrates how to instantiate, fit, and predict with the `RoPWR` model. It creates a simple dataset with two linear segments and fits a `RoPWR` model with two segments. The 'Clarabel' solver is specified, which is efficient for SOCP problems.

import numpy as np
from ropwr import RoPWR

# Generate some synthetic data
np.random.seed(0)
X = np.linspace(0, 10, 100).reshape(-1, 1)
y = np.where(X < 5, 2 * X, -X + 15).flatten() + np.random.normal(0, 1, 100)

# Initialize and fit the RoPWR model with 2 segments
model = RoPWR(n_segments=2, solver='Clarabel', random_state=42)
model.fit(X, y)

# Make predictions
y_pred = model.predict(X)

print(f"Predicted y for X[0]: {y_pred[0]:.2f}")
print(f"R^2 score: {model.score(X, y):.2f}")

view raw JSON →