This is a public, read-only report. Striff reads this repository's docs, turns each sentence that makes a claim about the code into a rule, and checks the rule against the code on the default branch. How this works
List<ValueVector> columns — the actual data, one vector per column
DataFrame has columns
Holds
RowIndex index — optional row labels (may be null)
DataFrame has index
Holds
describe() returns a new DataFrame with one row per column and columns: column, type, measure, count (non-null), mode, mean, std, min, 25%, 50%, 75%, max.
DataFrame has a describe method
Holds
sort() returns a new DataFrame with all rows reordered.
DataFrame has a sort method
Holds
If either frame has no RowIndex, falls back to merge().
DataFrame has a merge method
Holds
factorize() converts String columns into IntVector columns annotated with a NominalScale.
DataFrame has a factorize method
Holds
Both toArray() and toMatrix() convert the DataFrame to a dense numeric representation with optional bias (intercept) column and categorical encoding.
2 rules from this sentence
DataFrame has a toArray method
Holds
DataFrame has a toMatrix method
Holds
toString(from, to, truncate):
DataFrame has a toString method
Holds
JDBC types are mapped to SMILE types via StructType.of(ResultSetMetaData).
StructType has an of method
Holds
Enum value LEVEL, Meaning Integer level code (default), Output columns per category 1 — raw code value
CategoricalEncoder has LEVEL
Holds
Enum value DUMMY, Meaning Dummy / treatment encoding, Output columns per category k−1 binary columns (reference = first level)
CategoricalEncoder has DUMMY
Holds
Enum value ONE_HOT, Meaning Full one-hot encoding, Output columns per category k binary columns
CategoricalEncoder has ONE_HOT
Holds
Method new DataFrame(ValueVector...), Description Construct from column vectors Method new DataFrame(RowIndex, ValueVector...), Description With row index
DataFrame has a DataFrame method
Holds
Method DataFrame.of(double, String...), Description From 2-D double array
DataFrame has an of method
Holds
Method new StructType(StructField...), Description Construct from fields
StructType has a StructType method
Holds
Method field(int) / field(String), Description Get field by ordinal or name
StructType has a field method
Holds
Method indexOf(String), Description Ordinal of named field
StructType has an indexOf method
Holds
Method length(), Description Number of fields
StructType has a length method
Holds
Method names() / dtypes() / measures(), Description Field property arrays
3 rules from this sentence
StructType has a names method
Holds
StructType has a dtypes method
Holds
StructType has a measures method
Holds
Method add(StructField), Description Append a field (mutable)
StructType has an add method
Holds
Method rename(String, String), Description Rename a field (mutable)
StructType has a rename method
Holds
Constructor / method new StructField(name, dtype), Description Without measure Constructor / method new StructField(name, dtype, measure), Description With measure
Read.libsvm returns a SparseDataset<Integer> (not a DataFrame):
Read has libsvm
Holds
Write is a static-method interface;
Write is a contract
Holds
Method Write.csv(DataFrame, Path), Description CSV with default format Method Write.csv(DataFrame, Path, CSVFormat), Description CSV with explicit format
Method Write.arff(DataFrame, Path, String), Description ARFF with relation name
Write has an arff method
Holds
Method Write.object(Serializable), Description Serialize to a temp file (auto-deleted) Method Write.object(Serializable, Path), Description Serialize to a specific file
They all implement the interpolate(double... x) pattern (variadic double[] input) rather than a fixed-arity 2-D call.
9 rules from this sentence
ShepardInterpolation, ShepardInterpolation1D, ShepardInterpolation2D and 6 more have an interpolate method
Holdsall 9
Available radial basis functions (all in smile.math.rbf): Class GaussianRadialBasis, Formula φ(r) exp(−r²/2r₀²), Properties SPD Gram → Cholesky sensitive to scale r₀
GaussianRadialBasis is in smile.math.rbf
Holds
Variograms are used by KrigingInterpolation to model spatial dependence. All variograms implement Variogram (which extends smile.util.function.Function) via double f(double r).
KrigingInterpolation depends on Variogram
Holds
All variograms implement Variogram (which extends smile.util.function.Function) via double f(double r).
5 rules from this sentence
ExponentialVariogram implements Variogram
Holds
GaussianVariogram implements Variogram
Holds
PowerVariogram implements Variogram
Holds
SphericalVariogram implements Variogram
Holds
Variogram has a f method
Holds
Strongly-typed column vectors (IntVector, DoubleVector, StringVector, …) live in smile.data.vector.
3 rules from this sentence
IntVector is in smile.data.vector
Holds
DoubleVector is in smile.data.vector
Holds
StringVector is in smile.data.vector
Holds
A rich collection of distance and similarity metrics implementing the Distance<T> interface: Metric Euclidean, Class EuclideanDistance
EuclideanDistance implements Distance
Holds
Vector is a specialized DenseMatrix with either one row or one column.
Vector extends DenseMatrix
Holds
Its subclasses DenseMatrix64 (Float64) and DenseMatrix32 (Float32) store data in column-major off-heap MemorySegment with a cache-optimised leading dimension.
2 rules from this sentence
DenseMatrix64 extends DenseMatrix
Holds
DenseMatrix32 extends DenseMatrix
Holds
DenseMatrix implements java.io.Serializable.
DenseMatrix uses java.io.Serializable
Holds
Method int numClasses(), Description Number of distinct classes
2 rules from this sentence
Classifier has a numClasses method
Holds
Classifier has a classes method
Holds
Method int predict(T x), Description Hard prediction — returns the predicted class label
Method boolean isSoft(), Description true if predict(T, double[]) is available
Classifier has an isSoft method
Holds
Method boolean isOnline(), Description true if update(T, int) is available
Classifier has an isOnline method
Holds
Method void update(T x, int y), Description Online learning — update the model with a single new labelled sample
Classifier has an update method
Holds
Method int applyAsInt(T x), Description Alias for predict satisfies ToIntFunction<T>
Classifier has an applyAsInt method
Holds
Method double applyAsDouble(T x), Description Alias for score satisfies ToDoubleFunction<T>
Classifier has an applyAsDouble method
Holds
It stores the IntSet classes label encoder and provides default implementations of numClasses() and classes().
3 rules from this sentence
AbstractClassifier has classes
Holds
AbstractClassifier has a numClasses method
Holds
AbstractClassifier has a classes method
Holds
An extension of Classifier<Tuple> for models that are trained on DataFrame objects via a Formula.
DataFrameClassifier extends Classifier
Holds
Method Formula formula(), Description The formula used at training time
DataFrameClassifier has a formula method
Holds
Method StructType schema(), Description The schema of the feature columns
DataFrameClassifier has a schema method
Holds
Method int predict(Tuple x), Description Predicts using a named-column Tuple
DataFrameClassifier has a predict method
Holds
All tree-based models (DecisionTree, RandomForest, AdaBoost, GradientTreeBoost) implement DataFrameClassifier.
2 rules from this sentence
DecisionTree implements DataFrameClassifier
Holds
AdaBoost implements DataFrameClassifier
Holds
The static DataFrameClassifier.of(formula, classifier) helper adapts any Classifier<double[]> into a DataFrameClassifier, applying the formula's feature extraction automatically.
DataFrameClassifier has an of method
Holds
A utility record that maps arbitrary integer class labels to the contiguous range [0, k) required by internal array indexing.
ClassLabels is a concrete implementation
Holds
Field k, Type int, Description Number of classes
ClassLabels has a k field
Holds
Field classes, Type IntSet, Description Encoder classes.indexOf(label) → index, classes.valueOf(index) → label
ClassLabels has a classes field
Holds
Field y, Type int[], Description The original label array remapped to [0, k)
ClassLabels has a y field
Holds
Field ni, Type int[], Description Per-class sample counts
ClassLabels has a ni field
Holds
Field priori, Type double[], Description Estimated prior probabilities ni[i] / n
ClassLabels has a priori field
Holds
PlattScaling is used internally by OneVersusRest.fit and OneVersusOne.fit when the base classifier supports score().
2 rules from this sentence
OneVersusRest depends on PlattScaling
Holds
OneVersusOne depends on PlattScaling
Holds
LogisticRegression, RandomForest, GradientTreeBoost, AdaBoost, DecisionTree, MLP, and Maxent all expose a fit(…, Properties) overload and a nested Options record with toProperties() / of(Properties) for round-tripping:
4 rules from this sentence
LogisticRegression has a fit method
Holds
AdaBoost has a fit method
Holds
DecisionTree has a fit method
Holds
Maxent has a fit method
Holds
Clustering is a utility interface that defines the global OUTLIER constant and the shared Options record used by most partitioning algorithms.
2 rules from this sentence
Clustering is a contract
Holds
Clustering has OUTLIER
Holds
Partitioning is the base class for density-based algorithms (DBSCAN, HDBSCAN, MEC, DENCLUE) that do not produce centroids.
3 rules from this sentence
HDBSCAN extends Partitioning
Holds
MEC extends Partitioning
Holds
DENCLUE extends Partitioning
Holds
k — number of clusters found (not counting the outlier pseudo-cluster).
Partitioning has k
Holds
group[] — per-point cluster label;
Partitioning has group
Holds
CentroidClustering<T, U> is a record returned by all centroid-based algorithms.
CentroidClustering is a concrete implementation
Holds
Method k(), Returns int, Description Number of clusters
Ensemble models (RandomForest, GradientTreeBoost) implement TreeSHAP for feature importance.
RandomForest implements TreeSHAP
Holds
Uniform Regression<T> interface — a single predict(T x) for scalar output, plus batch, list, and dataset overloads;
2 rules from this sentence
Regression is a contract
Holds
Regression has a predict method
Holds
optional online update.
Regression has update
Holds
Method boolean online(), Description Returns true if the model supports incremental updates
Regression has an online method
Holds
Method double applyAsDouble(T x), Description Alias for predict satisfies ToDoubleFunction<T>
Regression has an applyAsDouble method
Holds
Ensemble average. The static factory method Regression.ensemble(models...) returns a model whose prediction is the arithmetic mean of all constituent models.
Regression has an ensemble method
Holds
Method Formula formula(), Description The formula used at training time
DataFrameRegression has a formula method
Holds
Method StructType schema(), Description The feature schema of the design matrix
DataFrameRegression has a schema method
Holds
Method double predict(Tuple x), Description Predicts from a named-column Tuple
DataFrameRegression has a predict method
Holds
All tree-based models (RegressionTree, RandomForest, GradientTreeBoost) and GLM, GAM, LinearModel implement DataFrameRegression.
5 rules from this sentence
RegressionTree implements DataFrameRegression
Holds
GLM implements DataFrameRegression
Holds
GAM implements DataFrameRegression
Holds
LinearModel implements DataFrameRegression
Holds
DataFrameRegression is a contract
Holds
The static helper DataFrameRegression.of(formula, model) adapts any Regression<double[]> into a DataFrameRegression, applying formula extraction automatically:
DataFrameRegression has an of method
Holds
Method double[] coefficients(), Description Fitted weights (last element is the intercept when included in the design)
LinearModel has a coefficients method
Holds
Method double intercept(), Description The intercept term
LinearModel has an intercept method
Holds
Method double RSquared(), Description Coefficient of determination R²
LinearModel has a RSquared method
Holds
Method double adjustedRSquared(), Description R² adjusted for the number of predictors
LinearModel has an adjustedRSquared method
Holds
Method double RSS(), Description Residual sum of squares
LinearModel has a RSS method
Holds
Method double error(), Description Root mean square error of the residuals (σ̂)
Each family is a Java interface (Bernoulli, Binomial, Poisson, Gaussian) exposing a static factory that returns a Model.
2 rules from this sentence
Bernoulli is a contract
Holds
Poisson is a contract
Holds
Family Bernoulli, Factory Bernoulli.logit(), Link logit, Response type Binary (0/1) outcomes
Bernoulli has a logit method
Holds
Family Poisson, Factory Poisson.log(), Link log, Response type Non-negative count data
Poisson has a log method
Holds
Both model types expose a predict(Tuple x) method that accepts a single data row:
2 rules from this sentence
ClassificationModel has a predict method
Holds
RegressionModel has a predict method
Holds
The "svm" and "gaussian-process" algorithms both accept a smile.svm.kernel / smile.gaussian_process.kernel property whose value is a short string parsed by MercerKernel.of(String).
MercerKernel has an of method
Holds
Both ClassificationModel and RegressionModel implement java.io.Serializable.
VisionModel.forward(BufferedImage...) automatically applies the model's associated Transform, so you never need to preprocess images manually.
2 rules from this sentence
VisionModel depends on Transform
Holds
VisionModel has a forward method
Holds
The EfficientNet constructor accepts an MBConvConfig[] array to define a custom architecture, giving fine-grained control over each inverted-residual stage:
EfficientNet depends on MBConvConfig
Holds
ImageNet is an interface with two 1000-element string arrays and a set of utility methods for mapping between class indices and human-readable labels:
ImageNet is a contract
Holds
GenAiChatModel implements smile.llm.LanguageModel for a future serve backend.
GenAiChatModel implements LanguageModel
Holds
Figure.toBufferedImage(width, height) renders to a BufferedImage with no display required.
Figure has a toBufferedImage method
Holds
Palette.jet(256, 0.8f).
Palette has a jet method
Holds
Palette.get(int index) returns a distinct color for the given group index (cycles through a built-in list of 10 visually distinct colous). It is used automatically by ScatterPlot.of(x, labels, mark) and similar factory methods.
ScatterPlot depends on Palette
Holds
Palette.get(int index) returns a distinct color for the given group index (cycles through a built-in list of 10 visually distinct colous).
Palette has a get method
Holds
It is used automatically by ScatterPlot.of(x, labels, mark) and similar factory methods.
ScatterPlot has an of method
Holds
All axis-level configuration is accessed via figure.getAxis(i), where i = 0 is the x-axis, i = 1 is the y-axis, and i = 2 is the z-axis (for 3-D figures).
Figure has a getAxis method
Holds
Legends are displayed automatically when a Plot has legends and figure.isLegendVisible() is true (the default).
Figure has an isLegendVisible method
Holds
Class Shape, Responsibility Abstract base — carries color knows how to paint(Renderer)
Shape has a paint method
Holds
Class Plot, Responsibility Shape subclass that also knows its data bounds ( getLowerBound / getUpperBound ) and optional legends/tooltips
3 rules from this sentence
Plot extends Shape
Holds
Plot has getLowerBound
Holds
Plot has getUpperBound
Holds
Class FigurePane, Responsibility Swing JPanel wrapping a Canvas with a toolbar
FigurePane depends on Canvas
Holds
It extends VegaLite, which provides the top-level properties shared by all specification types.
View extends VegaLite
Holds
view.data() returns a Data object that supports four source types.
View has a data method
Holds
view.mark(type) selects the geometric primitive and returns a Mark object for further configuration.
View has a mark method
Holds
view.encode(channel, field) binds a data field to a visual channel and returns a Field object for further refinement.
View has an encode method
Holds
view.transform() returns a Transform object that appends entries to the view-level transform array.