PredictorFunction¶

Status: Stable

documented, exercised by the test suite and/or worked examples, with no known limitations recorded.

Description¶

PredictorFunction[method, coefficients, featureCount]

is the fitted object Predict returns. Apply it to a feature vector to get a prediction, or to "Method", "Coefficients" or "FeatureCount" to read it. A one-feature model also accepts a bare scalar.

Examples (4)¶

Every input below was run against the current Mathilda build and its output recorded.

Applications (4)¶

Fit a predictor; it prints elided

In[1]:= p = Predict[{1 -> 2, 2 -> 4, 3 -> 6}]
Out[1]= PredictorFunction["LinearRegression", <>]

Apply it to a new input

In[2]:= p[10]
Out[2]= 20.0

Read its coefficients: {intercept, slope}

In[3]:= p["Coefficients"]
Out[3]= {0.0, 2.0}

And the method it was fitted with

In[4]:= p["Method"]
Out[4]= "LinearRegression"

Implementation notes¶

Definition. PredictorFunction[method, coefficients, featureCount] is the fitted predictor object that Predict and LinearModelFit return. Apply it to a feature vector to get a prediction, or to the strings "Method", "Coefficients" or "FeatureCount" to read it back; a one-feature model also accepts a bare scalar. It is Protected, prints elided (its parameters are derived), has no builtin of its own, and its docstring is in predict.c.

Representation. A positional inert head built by ml_make_model (src/ml/predict.c): a method-tag string ("LinearRegression", "NearestNeighbors"), a parameters slot whose shape the method chooses (a coefficient List for linear regression, the training matrix for nearest neighbours), the feature count, and an extra count. Applying the object routes through ml_model_apply, which confirms the head is PredictorFunction, reads the method tag, and evaluates the prediction; for linear regression the coefficient list is {intercept, w1, ...} and the result is its dot product with the feature vector.

Usage & limits. Produced by fitting, not written by hand. Predict implements only "LinearRegression" (the default); LinearModelFit returns the same object without regression diagnostics (RSquared, standard errors, ANOVA are not implemented). A collinear feature set has no unique fit, so Predict returns unevaluated rather than inventing one.

Attributes: Protected.

References¶

Notes & additional examples¶

Notes¶

PredictorFunction[method, coefficients, featureCount] is what Predict and LinearModelFit return. Apply it to a feature vector (or, for a one-feature model, a bare scalar) to get a prediction, or to "Method", "Coefficients" or "FeatureCount" to read it back. It prints elided because its parameters are derived. Predict implements only "LinearRegression", whose coefficient list is {intercept, w1, ...}; a collinear feature set has no unique fit and returns unevaluated.