ML Node Catalog

Load Trained ML Model from Path

2 in2 outSecurity 7/10

Load Model (Binary)

Load Trained ML Model from Path using fast binary format (Fory)

2 in2 outSecurity 7/10

Predict

Predict with Machine Learning Model

4 in1 outSecurity 6/10

Prediction Class/Label

1 in2 outSecurity unrated

Extract class_idx and label from predictions.

Save Model

Save Trained ML Model to Path

3 in1 outSecurity 7/10

Save Model (Binary)

Save Trained ML Model to Path using fast binary format (Fory)

3 in1 outSecurity 7/10

Teachable Machine

6 in2 outSecurity unrated

Image classification using Teachable Machine models.

Fit Novelty Detection (One-Class SVM)

Fit a One-Class SVM on normal observations only. Predictions flag whether a new row is an inlier (1) or an outlier (0).

6 in3 outSecurity 6/10

Train Classifier (AdaBoost)

Fit/Train an AdaBoost classifier using multi-class SAMME boosting over shallow Decision Trees. Each learner focuses on the rows its predecessors got wrong, so boosting usually beats a single tree on weak signal, but it is far more sensitive to label noise and outliers than Random Forest. Estimators is a maximum, not a guarantee: boosting stops early once a learner is no better than random guessing.

6 in3 outSecurity 6/10

Train Classifier (Decision Tree)

Fit/Train a Decision Tree classifier. Native multi-class support with interpretable rules.

7 in2 outSecurity 6/10

Train Classifier (K-Nearest Neighbours)

Fit a K-Nearest-Neighbours classifier. Non-parametric and instance based: the fitted model embeds a verbatim copy of the whole training set instead of learned coefficients, so every training row (and any personal data in it) travels with the model, is written into every saved model file and can be reconstructed by anyone holding it. Treat the model with the same care as the source table.

4 in2 outSecurity 3/10

Train Classifier (Logistic Regression)

Fit/Train a Logistic Regression classifier with L2 regularization. Handles binary and multi-class targets and yields interpretable coefficients plus calibrated probabilities. The solver expects features on a comparable scale - fit a Feature Scaler first if your columns have very different ranges.

8 in2 outSecurity 6/10

Train Classifier (Multinomial Naive Bayes)

Fit/Train a Multinomial Naive Bayes classifier, the standard baseline for text and other count data. Features must be non-negative counts or TF-IDF weights, which is what the Fit TF-IDF Vectorizer node produces. Native multi-class support and a single pass over the data.

3 in2 outSecurity 6/10

Train Classifier (Naive Bayes)

Fit/Train a Gaussian Naive Bayes classifier. Native multi-class support - no need for One-vs-All.

2 in2 outSecurity 6/10

Train Classifier (Random Forest)

Fit/Train a Random Forest classifier: many Decision Trees, each grown on a bootstrapped sample of the rows and a random subset of the features, combined by majority vote. Far more robust to overfitting than a single tree, at the price of interpretability. Model size and fit time grow linearly with Ensemble Size, so a forest of 500 trees costs roughly 500x a single tree.

8 in2 outSecurity 6/10

Train Classifier (SVM)

Fit/Train Support Vector Machines (SVM) for Multi-Class Classification

5 in2 outSecurity 6/10

Fit Clustering (Gaussian Mixture)

AI/ML/Clustering

Fit/Train a Gaussian Mixture Model. Soft clustering with per-component covariances and mixture weights, fitted by Expectation-Maximization.

10 in3 outSecurity 6/10

Train Clustering (DBSCAN)

AI/ML/Clustering

Fit/Train DBSCAN Density-Based Clustering

4 in3 outSecurity 6/10

Train Clustering (KMeans)

AI/ML/Clustering

Fit/Train KMeans Clustering

3 in2 outSecurity 6/10

K-Fold Split

Generate K train/test splits for cross-validation. Each fold uses (K-1)/K data for training and 1/K for validation, and runs the connected fold branch once per fold.

6 in4 outSecurity 6/10

Sample Dataset

Random sample N records or a ratio from a dataset

5 in2 outSecurity 6/10

Shuffle Dataset

Shuffle dataset rows randomly

3 in1 outSecurity 6/10

Split Dataset

Split a dataset into training and testing subsets

5 in1 outSecurity 6/10

Stratified Split

Split a dataset into training and testing subsets, keeping every class at its original proportion in both subsets

7 in1 outSecurity 6/10

Accuracy

Calculate classification accuracy by comparing predictions to actual values

4 in2 outSecurity 8/10

Confusion Matrix

Build confusion matrix and calculate precision, recall, and F1 score

4 in2 outSecurity 8/10

ROC-AUC & Log Loss

Threshold-free evaluation of a binary classifier: area under the ROC curve, log loss and the curve points. This is the payoff for Logistic Regression producing calibrated probabilities instead of bare class labels.

5 in4 outSecurity 8/10

Regression Metrics

Calculate MSE, RMSE, MAE, and R² for regression predictions

4 in2 outSecurity 8/10

Silhouette Score

Evaluate clustering quality: how much closer each sample sits to its own cluster than to the nearest other one (-1 to +1)

5 in4 outSecurity 8/10

Feature Importance

Extract per-feature importance from a Decision Tree, Random Forest or AdaBoost model

3 in4 outSecurity 10/10

Get Centroids

Extract cluster centroids from a trained KMeans model

2 in2 outSecurity 10/10

Get Coefficients

Extract coefficients and intercept from a trained Linear Regression model

2 in2 outSecurity 10/10

Model Info

2 in3 outSecurity unrated

Get general information about any ML model

Extract Keypoint

2 in5 outSecurity unrated

Extract a specific keypoint from a pose by index or name

Feature Extraction

4 in3 outSecurity unrated

Extract feature vectors from images using ONNX models

Feature Similarity

Compare two feature vectors using cosine similarity or L2 distance

Image Classification

7 in2 outSecurity unrated

Image Classification with ONNX-Models. Download models from: MobileNetV2 (https://github.com/onnx/models/tree/main/validated/vision/classification/mobilenet), SqueezeNet (https://github.com/onnx/models/tree/main/validated/vision/classification/squeezenet), ResNet (https://github.com/onnx/models/tree/main/validated/vision/classification/resnet), EfficientNet (https://github.com/onnx/models/tree/main/validated/vision/classification/efficientnet-lite4)

Load ONNX

2 in4 outSecurity unrated

Load ONNX Model from Path

Model Info

2 in4 outSecurity unrated

Get ONNX model metadata (inputs, outputs, shapes)

Object Detection

6 in2 outSecurity unrated

Object Detection in Images with ONNX-Models. Download models from: TinyYOLOv2 (https://github.com/onnx/models/tree/main/validated/vision/object_detection_segmentation/tiny-yolov2), YOLO (https://github.com/onnx/models/tree/main/validated/vision/object_detection_segmentation), SSD-MobileNet (https://github.com/onnx/models/tree/main/validated/vision/object_detection_segmentation/ssd-mobilenetv1)

Pose Estimation

5 in2 outSecurity unrated

Detect human poses and keypoints using ONNX models. Download models from: YOLOv8-Pose (https://docs.ultralytics.com/models/yolov8/), MoveNet (https://tfhub.dev/google/movenet/), HRNet (https://github.com/OAID/TengineKit)

Semantic Segmentation

Segment images into semantic classes using ONNX models. Download models from: DeepLabV3 (https://github.com/onnx/models/tree/main/validated/vision/object_detection_segmentation/duc), FCN (https://github.com/onnx/models/tree/main/validated/vision/object_detection_segmentation/fcn)

Session Info

1 in4 outSecurity unrated

Get information about a loaded ONNX session

Unload ONNX

Release ONNX model from cache to free memory

Audio to Mel Spectrogram

Convert audio to mel spectrogram for speech models

Load Audio

2 in4 outSecurity unrated

Load audio file for processing

Resample Audio

Resample audio to target sample rate

Trim Audio

Trim audio to speech segments from VAD

Voice Activity Detection

6 in3 outSecurity unrated

Detect speech segments in audio. Download Silero VAD model from: https://github.com/snakers4/silero-vad/raw/master/src/silero_vad/data/silero_vad.onnx

Batch Image Inference

AI/ML/ONNX/Batch

Run ONNX inference on multiple images in batches

6 in3 outSecurity unrated

Analyze Faces

4 in3 outSecurity unrated

Detect faces and extract embeddings, gender and age using a face_id analyzer

Compare Faces

4 in4 outSecurity unrated

Compare two face embeddings for similarity

Crop Faces

Crop detected faces from image

Face Detection

6 in3 outSecurity unrated

Detect faces in images. Download models from: UltraFace (https://github.com/onnx/models/tree/main/validated/vision/body_analysis/ultraface), RetinaFace (https://huggingface.co/arnabdhar/retinaface-onnx), SCRFD (https://huggingface.co/onnx-community/scrfd_10g_bnkps)

Face Embedding

Extract face embedding for recognition. Download models from: ArcFace (https://huggingface.co/onnx-community/arcface_torch/tree/main), FaceNet (https://huggingface.co/rocca/facenet-onnx)

Load Face Analyzer

11 in2 outSecurity unrated

Load a face_id analyzer (SCRFD detector + ArcFace embedder + gender/age). Weights are verified and cached when a session identity is first built; equivalent analyzers reuse process-wide sessions.

Unload Face Analyzer

Release a cached face analyzer and its three ONNX sessions. Equivalent analyzer handles share the same cache entry and are invalidated together.

8 in4 outSecurity unrated

Named Entity Recognition

AI/ML/ONNX/NLP

Extract named entities (persons, organizations, locations, dates, etc.) from text using ONNX models. Supports BERT, RoBERTa, and other transformer-based NER models with automatic tokenization. Download models from: BERT-base-NER (https://huggingface.co/dslim/bert-base-NER), Multilingual NER (https://huggingface.co/Davlan/bert-base-multilingual-cased-ner-hrl), spaCy NER (https://huggingface.co/spacy). Text longer than the model's window is split into overlapping chunks rather than truncated, so entities are found throughout a long document. Download tokenizer.json and config.json from the same model repository — config.json carries the id2label mapping that names the entity types and the sequence length the model accepts.

Zero-Shot NER (GLiNER)

AI/ML/ONNX/NLP

Extract entities for any labels you name at runtime, with no fixed label set and no retraining. Load a GLiNER ONNX export (e.g. https://huggingface.co/onnx-community/gliner_small-v2.1, gliner_multi-v2.1, gliner_medium_news-v2.1, gliner_multi_pii-v1, NuNER_Zero) plus the tokenizer.json from the same repository. For models with a fixed label set, use the Named Entity Recognition node instead.

9 in4 outSecurity unrated

Crop Text Regions

AI/ML/ONNX/OCR

Crop detected text regions from image for recognition

Text Detection

AI/ML/ONNX/OCR

Detect text regions in images. Download models from: CRAFT (https://huggingface.co/quocanh34/craft_text_detection_onnx), DBNet (https://huggingface.co/Xenova/dbnet_resnet50_onnx), EAST (https://www.dropbox.com/s/r2ingd0l3zt8hxs/frozen_east_text_detection.tar.gz)

Text Recognition

AI/ML/ONNX/OCR

Recognize text from cropped text regions. Download models from: CRNN (https://huggingface.co/Xenova/crnn_onnx), TrOCR (https://huggingface.co/microsoft/trocr-base-printed), PaddleOCR (https://huggingface.co/aapot/paddleocr-onnx)

Colorize Depth

AI/ML/ONNX/Vision

Convert depth map to rainbow-colored visualization

Depth Estimation

AI/ML/ONNX/Vision

Estimate depth from a single image using ONNX models. Download models from: MiDaS (https://github.com/isl-org/MiDaS/releases), DPT (https://huggingface.co/Intel/dpt-large/tree/main), Depth Anything (https://huggingface.co/depth-anything/Depth-Anything-V2-Small/tree/main)

4 in3 outSecurity unrated

Depth to Point Cloud

AI/ML/ONNX/Vision

Convert depth map to 3D point cloud coordinates

Ordinal Metrics

Evaluate predictions for an ordered target with distance-aware metrics. Plain accuracy is inadequate here: it treats "predicted high when the truth was medium" exactly as harshly as "predicted low", so a model that is reliably one level off scores like one that guesses. Quadratic weighted kappa is the standard headline metric because it weights every miss by how far off it was and corrects for chance agreement, but it answers only one of three questions: the linear kappa and the macro-averaged error say how far off the model is under a different cost structure and on the rare levels, while Kendall's tau-b and the Spearman correlation say whether it orders the rows correctly at all.

5 in12 outSecurity 8/10

Train Ordinal Model (Adjacent Category)

Fit/Train an ordinal model that compares each level with the one directly below it: `log( P(level k+1) / P(level k) ) = contrast_k + x . beta`. Its coefficients answer `what does one more unit of this feature do to my rating?` - `exp(coefficient)` is the factor on the odds of scoring one level higher rather than staying put, the same factor at every step. That is NOT what Train Ordinal Model (Proportional Odds) reports: a cumulative coefficient is the log odds ratio of everything AT OR BELOW a cut point against everything above it, pooling levels instead of comparing two neighbours. The same fitted number therefore means different things in the two families, and since one shared coefficient applies once per step here, the bottom-to-top effect is (levels - 1) times the per-step effect. Pick this for ratings, severity grades and Likert answers, where the question really is about one step; pick proportional odds when the question is about crossing a threshold (`does this case escalate past level 2?`). Fitted by penalized maximum likelihood over all levels jointly, so per-level probabilities are calibrated and the Predict node returns a confidence. Scale your features first with the Fit Feature Scaler node: this is a gradient fit, and unscaled columns make it converge slowly or not at all.

7 in5 outSecurity 6/10

Train Ordinal Model (Continuation Ratio)

Fit/Train a continuation-ratio model on an ORDERED target that is really a process that can halt. It fits K-1 sub-models, where sub-model k answers `given this row reached level k, did it STOP there?`, so the model describes a progression through the levels instead of placing cut points on a latent scale. Reach for it when the levels are genuinely sequential and each one had to be passed to get to the next: escalation tiers, disease stages, how far a signup funnel got, how far an incident escalated before it was contained. Each sub-model carries its own coefficient vector, so nothing assumes proportional odds, and the per-level probabilities are exact by the chain rule rather than differences of two fits. The cost is strictness: because each sub-model is conditioned on having reached its level, EVERY level must occur in the training data, middle ones included. Scale your features first with the Fit Feature Scaler node: these are gradient fits, and unscaled columns make them converge slowly or not at all.

8 in5 outSecurity 6/10

Train Ordinal Model (Frank & Hall)

Fit/Train an ordinal model by decomposition: the ordered target is cut K-1 times (`is the level above this cut?`) and each cut is handed to an ordinary binary classifier, with the predicted level read back as the number of cuts answered yes. This is the one ordinal trainer here that is not linear in the features, so reach for it when the boundary between levels bends in a way the Proportional Odds and Ridge trainers cannot follow. The price is that the K-1 sub-models are fitted independently: there is no single latent scale, no coefficient vector to read a direction off, and no calibrated per-level probabilities - use Proportional Odds when you need those. Every declared level must occur in the training data at the bottom and at the top of the ordering, otherwise a cut has only one class and cannot be fitted. A Random Forest base is the sturdiest choice and by far the costliest: each cut grows its own full forest, so training costs K-1 forests and the saved model carries every tree of every one of them.

4 in3 outSecurity 6/10

Train Ordinal Model (Neural CORAL/CORN)

Fit/Train a NEURAL ordinal model on a target whose levels are ORDERED (1 < 2 < ... < 5, or low < medium < high). This is the only trainer in the catalog that is BOTH non-linear in the features AND yields calibrated, rank-consistent per-level probabilities: Frank & Hall is non-linear but votes with K-1 independent classifiers and therefore carries no probability model, while every other ordinal node here is linear in the features. A small network feeds one of two rank-consistent heads, CORAL or CORN, and both are built so that P(y > k) can never rise with k for ANY parameter values — so the level probabilities are non-negative and sum to 1 with nothing patched up afterwards. THE HONEST LIMIT: leave Hidden Layers EMPTY and CORAL becomes exactly Train Ordinal Model (Proportional Odds) with Loss = AllThreshold and Margin = Logistic, and CORN becomes exactly Train Ordinal Model (Continuation Ratio) — the same objective in the same parameters. The hidden layers are the entire contribution, so if your problem is linear in the features prefer those nodes: convex objective, no seed dependence, readable coefficients, better tested. Reach for this one when the level is genuinely not monotone in the features (a boundary that bends back on itself, which no linear ordinal model can represent at all). Two costs come with the network: it has far more parameters than a linear model and so needs far more rows — check the Architecture output — and the objective is not convex, so the Seed changes the fit. Scale your features first with the Fit Feature Scaler node; unscaled columns make this converge slowly or not at all.

11 in5 outSecurity 6/10

Train Ordinal Model (Proportional Odds)

Fit/Train a proportional-odds model on a target whose levels are ORDERED (1 < 2 < ... < 5, or low < medium < high). Use this instead of a classifier, which treats the levels as unrelated names and so counts predicting `low` for `high` as no worse than predicting `medium`. Use it instead of a regressor, which treats the levels as real numbers and so invents distances the levels do not carry (`high` is not exactly twice `medium`). The model learns one coefficient vector plus ordered cut points, which keeps predictions monotone in the score and, under the default loss, yields calibrated per-level probabilities. Link Function, Loss and Margin widen it to the whole threshold-model family, up to support vector ordinal regression, while Free Features relaxes the shared coefficient into one slope per cut point. Scale your features first with the Fit Feature Scaler node: this is a gradient fit, and unscaled columns make it converge slowly or not at all.

11 in6 outSecurity 6/10

Train Ordinal Model (Ridge)

Fit/Train an ordinal model the cheap way: ridge-regress the level rank on the features, then cut the score at thresholds learned from the training distribution instead of rounding it. Closed-form, so it stays fast exactly where the proportional-odds model gets expensive - many levels, many features, or when you just want a quick ordinal baseline to beat. It also degrades gracefully when the proportional-odds assumption does not hold. Unlike the proportional-odds model it yields no probabilities: you get the predicted level and the latent score behind it, nothing calibrated.

4 in4 outSecurity 6/10

Apply Transform

AI/ML/Preprocessing

Apply a fitted transformer (Feature Scaler, TF-IDF) to a table, writing one vector per row. A Feature Scaler replays the exact offsets and scales learned at fit time, so applying it to train and test gives both the same statistics. TF-IDF is different: linfa recomputes the inverse document frequencies from the table being transformed, so vectors are only comparable within a single Apply Transform run.

4 in1 outSecurity 6/10

Fit Feature Scaler

AI/ML/Preprocessing

Learn per-feature offsets and scales from a training table. Distance- and gradient-based models (Logistic Regression, Elastic Net, SVM, KNN, Gaussian Mixture) only behave when their features share a scale.

5 in4 outSecurity 7/10

Fit TF-IDF Vectorizer

AI/ML/Preprocessing

Learn a vocabulary from a text column and turn documents into numeric vectors weighted by term frequency times inverse document frequency. Feed the fitted vectorizer to Apply Transform to vectorize a column, then train a classifier such as Multinomial Naive Bayes on the result. Tokenization always uses the built-in regex tokenizer, because a custom tokenizer function cannot be persisted and would make the saved model unloadable.

10 in3 outSecurity 6/10

PCA Reduction

AI/ML/Reduction

Principal Component Analysis for dimensionality reduction

3 in2 outSecurity 6/10

t-SNE Reduction

AI/ML/Reduction

t-Distributed Stochastic Neighbor Embedding. Projects high-dimensional vectors into 2-3 dimensions for visualization and writes the embedding back into the source table. t-SNE is transductive, so it produces no reusable model.

6 in1 outSecurity 6/10

Train Regression (Linear)

Fit/Train Linear Regression Model

2 in2 outSecurity 6/10

Train Regressor (GLM / Tweedie)

Fit/Train a Generalized Linear Model. Pick the distribution that matches the target: Normal for unbounded values, Poisson for counts, Gamma for positive skewed amounts, Inverse Gaussian for heavy tails.

8 in2 outSecurity 6/10

Train Regressor (K-Nearest Neighbours)

Fit a K-Nearest-Neighbours regressor that averages the target of the nearest training rows. Non-parametric and instance based: the fitted model embeds a verbatim copy of the whole training set instead of learned coefficients, so every training row (and any personal data in it) travels with the model, is written into every saved model file and can be reconstructed by anyone holding it. Treat the model with the same care as the source table.

4 in2 outSecurity 3/10

Train Regressor (Ridge/Lasso/ElasticNet)

Fit/Train a penalized linear regression model. Ridge shrinks all coefficients, Lasso drives irrelevant ones to exactly zero (feature selection), Elastic Net mixes both.

8 in3 outSecurity 6/10

Train Regressor (SVM)

1 in1 outSecurity unrated

Fit/Train a Support Vector Regressor. Learns non-linear targets through a kernel, with epsilon-SVR or nu-SVR.

9 in3 outSecurity 6/10

Prediction Score

AI/ML/Teachable Machine

Extract score from predictions.

Auto Classifier

Automatically finds the best classification model. Cross-validates Naive Bayes, Decision Tree, Logistic Regression, Random Forest and SVM, then retrains the winner on the full dataset. The reported Best Model Type can be fed straight into Grid Search to tune it further.

7 in4 outSecurity 6/10

Auto Ordinal

Automatically finds the best model for a target whose levels are ORDERED (1 < 2 < ... < 5, or low < medium < high). Cross-validates the ordinal families - Proportional Odds and Ordered Probit, the all-threshold model and its support-vector form, Ordinal Ridge, Continuation Ratio and Adjacent Category, plus an optional rank-consistent neural family that is off by default because it costs far more than all the others combined - on identical folds, ranks them by an ordinal metric that knows how far a miss was, then retrains the winner on the full data. Use this rather than Auto Classifier, which resolves the target without its order and ranks by accuracy or macro-F1, scoring a five-level miss exactly like a one-level one. Every candidate here is a gradient or a least-squares fit on the raw columns, so scale your features with the Fit Feature Scaler node first: unscaled columns change which family wins, not just how fast it converges.

12 in5 outSecurity 6/10

Grid Search

Exhaustive search over parameter combinations with cross-validation. Returns the best parameters found. Model Type accepts the same names the Auto Classifier reports as its best model, so the two nodes chain directly.

4 in3 outSecurity 6/10

Ordinal Grid Search