Question 1

Does MLdeck upload my CSV data to any server?

Accepted Answer

During normal browser training flows, MLdeck processes raw CSV training data locally in your browser using Pyodide WebAssembly inside a WebWorker thread. Network inspection should show no raw CSV upload during the training process. Backend services are used only for app, account, or control-plane features.

Question 2

Can I export trained ML models as ONNX?

Accepted Answer

MLdeck attempts ONNX export for the winning scikit-learn pipeline when supported by the browser runtime converter. When conversion succeeds, the ONNX graph can include preprocessing metadata and nodes for imputation, P1/P99 clipping, StandardScaler normalization, and OneHotEncoding. Some estimators are disclosed as unavailable instead of being silently substituted.

Question 3

Does MLdeck require Python or any software installation?

Accepted Answer

No installation required for local browser training. The scikit-learn Python runtime is compiled to WebAssembly via Pyodide and runs in a WebWorker thread. Users with no Python experience can train, evaluate, and export portable ML models by uploading a CSV file in any modern browser.

Question 4

What machine learning algorithms does MLdeck support?

Accepted Answer

MLdeck supports 10+ browser-friendly algorithm families, including SGD-style linear classifiers and regressors, Passive Aggressive, Naive Bayes, Decision Tree, Random Forest, Extra Trees, HistGradientBoosting, K-Nearest Neighbors, SVC/SVR options, and lightweight MLP models. Some labels describe browser-safe SGD variants rather than classical batch estimators.

Question 5

Can MLdeck's exported ONNX models run completely offline?

Accepted Answer

When exact ONNX conversion succeeds, the exported ONNX file is designed as a portable model graph that does not require a connection to MLdeck for inference. Validate preprocessing parity and runtime behavior before using it in an air-gapped Docker container, Python application, edge device, or other deployment target.

Question 6

How does MLdeck handle preprocessing and feature engineering?

Accepted Answer

MLdeck applies a deterministic 6-step preprocessing chain: (1) Median/mode imputation for missing values, (2) P1–P99 percentile clipping for outlier control, (3) StandardScaler normalization, (4) OneHotEncoding for categorical columns with unknown-value vocabulary, (5) Optional polynomial feature expansion, (6) Mutual information feature selection. When exact ONNX conversion succeeds, export attempts to include the active preprocessing chain in the model graph.

Question 7

What is the Docker inference package that MLdeck exports?

Accepted Answer

MLdeck auto-generates a Docker-style API package for runtime testing. The package includes preprocessing metadata, encoding schema, and exact ONNX artifacts when available. Run the generated container locally to test predictions and review behavior in your target environment.

Question 8

What is prequential evaluation in MLdeck?

Accepted Answer

Prequential evaluation (test-then-train protocol) is an online machine learning evaluation method used by MLdeck for large CSV files. The model is evaluated on each data chunk before being trained on it, producing streaming performance estimates without requiring a separate holdout set. Use external validation for production decisions.

Question 9

Is MLdeck free to use?

Accepted Answer

Yes. MLdeck's core browser training, evaluation, and supported export attempts are free to start. Since training runs on your local device via WebAssembly, there are no cloud compute charges, GPU rental costs, or server-side infrastructure fees.

Question 10

What is InferenceLab in MLdeck?

Accepted Answer

InferenceLab is MLdeck's browser-native model validation environment. When an ONNX graph is available, it loads the export using ONNX Runtime Web and runs single-row predictions directly in the browser, helping users check preprocessing fidelity and compare exported model behavior before external deployment.

Question 11

What is ONNX and why does MLdeck use it?

Accepted Answer

ONNX (Open Neural Network Exchange) is an open standard for representing machine learning models, supported by Microsoft, Meta, Intel, NVIDIA, and AWS. MLdeck uses ONNX because it is designed for portable inference across ONNX Runtime environments such as Python, C++, Java, .NET, Rust, JavaScript, and edge hardware, with schema and parity-review metadata included in export packages.

Question 12

How does MLdeck compare to Google AutoML, AWS SageMaker, or Azure AutoML?

Accepted Answer

Unlike Google AutoML, AWS SageMaker, and Azure AutoML, MLdeck does not require uploading raw CSV training data to cloud infrastructure during normal browser training flows. Training runs locally in the browser. MLdeck exports standard ONNX format and generates Docker-style packages for self-hosted runtime testing. Local browser training has no cloud compute billing.

MLdeck browser-local AutoML