Predicted R-Squared (r2, r^2) Calculation in `python`

stats.py

def press_statistic(y_true, y_pred, xs):
    """
    Calculation of the `Press Statistics <https://www.otexts.org/1580>`_
    """
    res = y_pred - y_true
    hat = xs.dot(np.linalg.pinv(xs))
    den = (1 - np.diagonal(hat))
    sqr = np.square(res/den)
    return sqr.sum()

def predicted_r2(y_true, y_pred, xs):
    """
    Calculation of the `Predicted R-squared <https://rpubs.com/RatherBit/102428>`_
    """
    press = press_statistic(y_true=y_true,
                            y_pred=y_pred,
                            xs=xs
    )

    sst  = np.square( y_true - y_true.mean() ).sum()
    return 1 - press / sst
 
def r2(y_true, y_pred):
    """
    Calculation of the unadjusted r-squared, goodness of fit metric
    """
    sse  = np.square( y_pred - y_true ).sum()
    sst  = np.square( y_true - y_true.mean() ).sum()
    return 1 - sse/sst

Could you please give us an example to calculate predicted_r2? Many thanks

If xs is the independent variable, it should be a 1-dimensional array, which causes an error in the line hat = xs.dot(np.linalg.pinv(xs)) that reads LinAlgError: 1-dimensional array given. Array must be at least two-dimensional.

Am I missing something here?

Not quite sure about press_statistic, can you link how the calculation in press_statistic actually ends up with PRESS?

The link you shared in docstring is now invalid.

Thanks.

Okay, I compared your code:

def press_statistic(y_true, y_pred, xs):
    """
    Calculation of the `Press Statistics <https://www.otexts.org/1580>`_
    """
    res = y_pred - y_true
    hat = xs.dot(np.linalg.pinv(xs))
    den = (1 - np.diagonal(hat))
    sqr = np.square(res/den)
    return sqr.sum()

with the original R implementation:

PRESS <- function(linear.model) {
  #' calculate the predictive residuals
  pr <- residuals(linear.model)/(1-lm.influence(linear.model)$hat)
  #' calculate the PRESS
  PRESS <- sum(pr^2)
  
  return(PRESS)
}

And can confirm the results are the same:

Python:

R:

---

Thanks.

However, you should make it clear that your code works only for models with an intercept. If you force the intercept to 0 (no intercept), your code will not work due to incorrect SST calculation:

You calculated the centered total sum of squares (SST):

sst  = np.square( y_true - y_true.mean() ).sum()

But the original R implementation calculates the uncentered total sum of squares:

+     #' Use anova() to get the sum of squares for the linear model
+     lm.anova <- anova(linear.model)
+     #' Calculate the total sum of squares
+     sst <- sum(lm.anova$'Sum Sq')

If you use the sample data in my comment above:

• Your sst (centered) = 640.9
• Original R implementation sst (uncentered) = 5437