std.ml_full

Erweiterte Machine-Learning-Algorithmen und Modelle: Deep Learning Layers, Convolutional Networks, Rekurrente Netze (LSTM), Gradient Boosting und Ensemble-Methoden. Aufbauend auf std.ml mit schwergewichtigen Implementierungen für Produktionseinsatz.

→ std.ml · Standard Library · std.graphdb

 
Die Doppel von ExpF64, LogF64 und SqrtF64 sind entfernt (#1572 behoben). Die Unit importiert jetzt std.math; wer sie mitimportiert, bekommt dieselben Werte wie überall sonst — mit 1.1.2E nachgemessen:

LogF64(100) = 4.605170
ExpF64(1)   = 2.718281


Vorher lieferte LogF64 hier immer 0 und ExpF64 eine grobe Näherung, ohne jede Meldung — und zwar im ganzen Programm, weil die Namen im flachen Namensraum kollidierten.

Typen

DecisionTreeNode (struct)

Feld Typ
featureIdx int64
threshold f64
left int64
right int64
classLabel int64
nSamples int64
impurity f64
featureImportance f64

Funktionen

Signatur Sichtbarkeit Beschreibung
SquareF64(x: f64): f64 pub —
SqrtF64(x: f64): f64 pub SSE2-optimized square root (uses sqrtsd when available) Currently falls back to Newton's method
SqrtF64SSE(x: f64): f64 pub —
ExpF64(x: f64): f64 pub Exponential function - simple approximation exp(x) ≈ 1 + x for small x, use lookup for common values
SigmoidF64(z: f64): f64 pub Sigmoid function: σ(z) = 1 / (1 + e^(-z)) Use piecewise approximation
LogF64(x: f64): f64 pub Log (natural logarithm) - inverse of exp
LinearRegressionInit(): void pub Initialize model - with small non-zero weights
LinearRegressionWeight(): f64 pub Get weight
LinearRegressionBias(): f64 pub Get bias
LinearRegressionLoss(): f64 pub Get last loss
LinearRegressionEpochs(): int64 pub Get epochs
LinearRegressionPredict(x: f64): f64 pub Predict single value (linear)
LogisticRegressionInit(): void pub Initialize logistic regression model
LogisticRegressionWeight(): f64 pub Get weight
LogisticRegressionBias(): f64 pub Get bias
LogisticRegressionLoss(): f64 pub Get last loss (binary cross-entropy)
LogisticRegressionAccuracy(): f64 pub Get accuracy
LogisticRegressionEpochs(): int64 pub Get epochs
LogisticRegressionPredictProb(x: f64): f64 pub Predict probability (sigmoid output) Returns value between 0 and 1
LogisticRegressionPredict(x: f64): int64 pub Predict class (binary: 0 or 1) Returns 0 if prob < 0.5, else 1
LinearRegressionFitDataFrame(df: int64, xCol: int64, yCol: int64): void pub df: DataFrame pointer xCol: column name for feature (int64 pointer) yCol: column name for target (int64 pointer)
LinearRegressionDataPointCount(): int64 pub Get number of data points from last DataFrame fit
LogisticRegressionFitDataFrame(df: int64, xCol: int64, yCol: int64, learningRate: f64, epochs: int64): void pub DataFrame-based Logistic Regression (arbitrary data points) Train Logistic Regression from DataFrame
LinearRegressionFitArrays(xData: int64, yData: int64, n: int64): void pub xData: pointer to f64 array (x values) yData: pointer to f64 array (y values) n: number of data points
LogisticRegressionFitArrays(xData: int64, yData: int64, n: int64, learningRate: f64, epochs: int64): void pub Array-based Logistic Regression (arbitrary data points) Train Logistic Regression from f64 arrays
KNNInit(k: int64): void pub Initialize k-NN model k: number of neighbors (usually odd: 1, 3, 5, 7…)
KNNGetK(): int64 pub Get k value
KNNGetNSamples(): int64 pub Get number of training samples
KNNFit(features: int64, labels: int64, n: int64, nFeatures: int64): void pub labels: pointer to labels array (n labels) n: number of training samples nFeatures: number of features per sample
KNNDistance(a: int64, b: int64, n: int64): f64 pub Euclidean distance between two feature vectors
KNNDistanceSSE(a: int64, b: int64, n: int64): f64 pub SSE-optimized Euclidean distance (future: uses AVX)
KNNManhattanDistance(a: int64, b: int64, n: int64): f64 pub Manhattan distance (L1)
KNNFindNeighbors(query: int64, k: int64): int64 pub Find k nearest neighbors (indices array)
KNNMajorityVote(neighborIndices: int64, k: int64): int64 pub Majority vote
KNNWeightedVote(neighborIndices: int64, query: int64, k: int64): int64 pub Weighted majority vote
KNNPredict(query: int64): int64 pub Predict class
KNNPredictProba(query: int64, nClasses: int64): int64 pub Predict probabilities
KNNCrossValidation(features: int64, labels: int64, n: int64, nFeatures: int64, k: int64, folds: int64): f64 pub Cross-validation
KNNScore(features: int64, labels: int64, n: int64): f64 pub Score on training set
KMeansInit(k: int64): void pub Cluster centroid vector Note: stored as contiguous f64 array at km_centroids Initialize k-Means
KMeansGetK(): int64 pub Get number of clusters
KMeansGetIter(): int64 pub Get current iteration
KMeansGetInertia(): f64 pub Get inertia (sum of squared distances)
KMeansInitCentroids(features: int64, n: int64, nFeatures: int64, k: int64): void pub Initialize centroids using k-means++ algorithm
KMeansFit(features: int64, n: int64, nFeatures: int64, k: int64): void pub n: number of samples nFeatures: number of features k: number of clusters
KMeansPredictCluster(point: int64): int64 pub Get cluster assignment for a point
KMeansGetAssignments(): int64 pub Get all cluster assignments
KMeansGetCentroids(): int64 pub Get cluster centroids
KMeansGetClusterSize(clusterIdx: int64): int64 pub Get cluster size
KMeansGetCentroid(idx: int64, featureIdx: int64): f64 pub Get centroids as array (for visualization)
KMeansSilhouette(features: int64, n: int64, nFeatures: int64): f64 pub Calculate silhouette score (quality metric)
KMeansElbow(features: int64, n: int64, nFeatures: int64, maxK: int64): int64 pub Elbow method: calculate inertia for different k values
KMeansMiniBatchInit(k: int64, batchSize: int64): void pub Initialize mini-batch k-means
KMeansMiniBatchFit(features: int64, n: int64, nFeatures: int64, k: int64): void pub Fit mini-batch k-means (faster approximation)
NaiveBayesInit(nClasses: int64, vocabSize: int64): void pub Initialize Naive Bayes
NaiveBayesFit(documents: int64, docLengths: int64, labels: int64, n: int64): void pub docLengths: lengths of each document labels: class label for each document n: number of documents
NBLog(x: f64): f64 pub Log function approximation
NaiveBayesPredict(docWords: int64, docLen: int64): int64 pub Predict class for a document
NaiveBayesPredictProba(docWords: int64, docLen: int64): int64 pub Predict probabilities for all classes
NaiveBayesGetWordCount(classIdx: int64, wordIdx: int64): f64 pub Get word count for class
NaiveBayesGetClassCount(classIdx: int64): int64 pub Get class count
NaiveBayesScore(documents: int64, docLengths: int64, labels: int64, n: int64): f64 pub Score on training data
NaiveBayesBernoulliInit(nClasses: int64, vocabSize: int64): void pub Initialize Bernoulli Naive Bayes
NaiveBayesBernoulliFit(documents: int64, docLengths: int64, labels: int64, n: int64): void pub Bernoulli NB training
NaiveBayesBernoulliPredict(docWords: int64, docLen: int64): int64 pub Bernoulli NB prediction (with complement)
DecisionTreeInit(maxDepth: int64, criterion: int64): void pub Initialize Decision Tree
DTGiniImpurity(labels: int64, n: int64): f64 pub Calculate Gini impurity
DTEntropyImpurity(labels: int64, n: int64): f64 pub Calculate Entropy impurity
DTMSEImpurity(values: int64, n: int64): f64 pub Calculate MSE impurity (for regression)
DecisionTreePredict(featureVector: int64): int64 pub Predict using Decision Tree
DecisionTreePredictProba(featureVector: int64, nClasses: int64): int64 pub Predict probabilities
DecisionTreeScore(features: int64, labels: int64, n: int64): f64 pub Score Decision Tree
DecisionTreeGetDepth(): int64 pub Get tree depth
DecisionTreeGetNLeafs(): int64 pub Get number of leaf nodes (approximation)

Signaturen automatisch aus std/ml_full.lyx erzeugt (Stand 2026-08-02, lyxc 1.0.21A). Beschreibungen stammen aus den Quellkommentaren und sind teilweise unvollständig.