Python wrapper around R's lovely smooth.spline

import matplotlib.pyplot as plt import numpy as np from pysmoothspl import SmoothSpline n = 1000 x = np.arange(n).astype(np.float) y = 100 + 50 * np.sin(2 * np.pi * x / 365.0) + np.random.normal(0, 25, n) spl = SmoothSpline(spar=0.2).fit(x, y) yhat = spl.predict(x) spl_smoother = SmoothSpline(spar=0.5).fit(x, y) yhat_smoother = spl_smoother.predict(x) fig, ax = plt.subplots(1, figsize=(16, 9)) ax.plot(x, y, 'ro', label='Points') ax.plot(x, yhat, 'g-', label='Less smooth') ax.plot(x, yhat_smoother, 'b-', label='Smoother') ax.legend() fig.show()

You will need Cython (unless/until this repo includes the *.c files) and numpy.

There are two Conda environment.yaml files to help guide the installation:

1. environment.yaml contains just the needed packages for installation
2. tests/environment.yaml contains the packages needed to install and test this package

This package uses py.test to run tests, and you will also need pandas:

py.test tests/

If rpy2 is installed, the tests will actually calculate the expected values of calculations using the copy of R it is installed against. For users without rpy2, the test data also contains "cached" answers calculated using R code.

• Object oriented sklearn-esque estimator
• For function inputs: typed memory views > NumPy buffers
• Kill the GIL
• CI service tests
  • Code coverage checks with CI
• Check output against literal R output via rpy2
• Implement more features
  • Cross validation
  • Derivatives (hint: check bvalue code for commented out)
• Work back from fresh copy of R code...