CM_ATIZAZUI by ATIZAZ — A complete modern WinForms UI library. 17 fully animated controls with glow, particles, text effects, gradient, and full customization. Works on .NET Framework 4.6.2 and .NET 6/7/8.
A SkiaSharp Canvas for PeachDrawing.Core - draw shapes, gradients, clips, layers, text and images with Skia, hardware accelerated when the surface you supply is, into a canvas you already have, with clickable links and no PDF, HTML or CSS involved.
Partial differential equations: heat, wave and Poisson solvers, finite element method, sparse matrices, conjugate gradients.
Neural network library for C#, written from scratch: Dense, Conv, MaxPool and RNN layers, Adam/Momentum/SGD, L2 and dropout, data-parallel training, gradient-checked and verified against PyTorch.
Retained vector scene (shapes, paths, texts, groups, gradients), deterministic SVG writer and text measurement with the metrics of Helvetica or any OpenType font. The drawing model of Meziantou.Framework.Charts and Meziantou.Framework.Diagrams.
ASP.NET Core middleware that self-adjusts its request concurrency limit from live CPU and latency-gradient signals (Vegas/AIMD-style), rejecting over-limit requests with 503 + Retry-After instead of queuing them. Zero-config defaults derived from ProcessorCount; container/cgroup CPU quotas respected.
Aquaculture AI: growth, mortality, disease and harvest prediction, feed optimisation policy, gradient-boosted DO forecasting and farm-data growth learning, ML.NET and ONNX Runtime backends and an LLM aquaculture agent with grounded tool calls.
The ML.NET integration for OptunaSharp (a .NET port of Optuna): ML.NET-backed forest parameter-importance evaluators over the Microsoft.ML.FastTree runtime. A faithful FastForest backend — FastForestFanovaImportanceEvaluator (fANOVA — Hutter et al. 2014) and FastForestMeanDecreaseImpurityImportanceEvaluator (MDI) — plus EXPERIMENTAL / off-label gradient-boosted-tree backends (ML.NET FastTree and FastTreeTweedie): FastTree/FastTreeTweedie {Fanova,MeanDecreaseImpurity}ImportanceEvaluator. The canonical FanovaImportanceEvaluator / MeanDecreaseImpurityImportanceEvaluator ship in Core as a pure-managed, bit-exact port of scikit-learn's RandomForestRegressor (value-parity with Python Optuna); FastForest fits that same bagging model with ML.NET (ranking parity — values diverge, not tree-identical to scikit-learn), while the boosters fit the ensemble residual rather than the objective, so fANOVA/MDI over them is off-label — a non-standard quantity with NO value-parity to Optuna (weak ranking agreement at best). Prefer the canonical Core evaluators when Optuna parity matters. Also ships the ML.NET backend for the Core OptunaSearchCV (the Phase-6 optuna.integration analog): MLNetEstimator, the ICrossValidationEstimator that cross-validates an arbitrary ML.NET pipeline — an adapter (validated by usage, not bit-parity), since ML.NET is not scikit-learn. (OptunaSearchCV itself, the ICrossValidationEstimator seam, and the byte-exact native scikit-learn backend all live in Core — the latter under namespace OptunaSharp.Integration.Sklearn.) The framework-agnostic PruningCallback — for driving pruning from your own training loop — lives in Core (OptunaSharp) as the base that framework integration packages extend. LightGBM lives in the separate OptunaSharp.Integration.LightGbm package (distinct heavy native runtime + Optuna tuner identity). Part of the SciSharp STACK.
CPU dense neural networks with deterministic initialization, mini-batch gradient descent, analytic gradients and validated JSON model persistence.
A CoreGraphics Canvas for PeachDrawing.Core - draw shapes, gradients, clips, layers, text and images with Apple's CoreGraphics into a caller's CGContext or an offscreen bitmap on macOS and iOS, with Core Image effects on the GPU, clickable links, and no PDF, HTML or CSS involved.
OptunaSharp's LightGBM integration (a .NET port of Optuna's optuna.integration.lightgbm / optuna_integration.lightgbm), backed by ML.NET's native LightGBM. LightGBMTuner / LightGBMTunerCV (and the train() drop-in): the stepwise LightGBM hyperparameter tuner — feature_fraction, num_leaves, bagging, regularization, min_child_samples in a fixed order — whose Optuna-driven search (GridSampler / TPESampler) is a faithful, oracle-validated port, with training bridged to ML.NET's native LightGBM (a Phase 6 adapter; inject ILightGbmTrainer for a different engine). Also ships EXPERIMENTAL / off-label gradient-boosted-tree importance backends (LightGbmFanovaImportanceEvaluator, LightGbmMeanDecreaseImpurityImportanceEvaluator): Optuna's importance uses only scikit-learn's RandomForestRegressor (bagging), so fANOVA/MDI over a LightGBM booster is a non-standard quantity with NO value-parity to Optuna — prefer the canonical evaluators in Core. Also ships LightGBMPruningCallback (the Phase-6 port of optuna_integration.lightgbm.LightGBMPruningCallback): the metric / valid-name / report-interval lookup + the study-direction consistency check + report → should-prune → TrialPruned, fed the per-iteration evaluation log from your loop (ML.NET's LightGBM exposes no managed per-iteration hook). Part of the SciSharp STACK.
BoTorchSampler for OptunaSharp — a native OptunaSharp BaseSampler that drives the real BoTorch (PyTorch + GPyTorch) library in-process via Python.NET (pythonnet). A faithful C# port of optuna_integration's BoTorchSampler glue (sampler protocol + candidates_func selection) whose numerics are direct botorch calls, so it is byte-exact with BoTorch. You pick the acquisition function with the pythonnet-free BoTorchCandidatesFunc enum: LogEI, qEI, qLogEI and qNEI for one objective; qEHVI, EHVI, qNEHVI, qParEGO and qLogEI-ParEGO for several; the knowledge-gradient qKG and qHVKG; constrained optimization through Trial.SetConstraint. Does NOT require the Python optuna_integration package — install only `botorch`. The .NET analog of using optuna_integration.BoTorchSampler, built on OptunaSharp.Interop.pythonnet. Part of the SciSharp STACK.
Blazor component (OsgCanvas) that runs an OpenSceneGraph.Net viewer in .NET and renders it with a WebGL2 renderer: lighting (8 lights), textures, fog, sky gradient, transparency, points/lines, HUD cameras, render-to-texture, shadow maps and bloom/HDR post-processing. Works with Blazor WebAssembly and Blazor Server. Created by Gravicode Studios, led by Kang Fadhil.
Cross-platform skeleton loading for .NET MAUI with theme-aware gradients and reduced-motion support.
Gravicode.CAD simulation: linear static finite element analysis — 2D/3D trusses, plane-stress solids (CST elements) with von Mises results, steady-state 2D heat conduction, a triangle mesher for profiles, sparse CSR matrices with a Jacobi-preconditioned conjugate-gradient solver, and colour-mapped result models.
Custom WinForms ProgressBar control with three-color gradient rendering, configurable border and aligned text overlay. Based on taancer/aProgressBar (https://github.com/taancer/aProgressBar).
Machine learning for Electrical.Net: load forecasting (gradient boosting, neural nets), fault location and classification, topology identification from measurements, and reinforcement-learning Volt/VAR agents, built on GravicodeScience (GraviLearn), ML.NET and Gravicode.RLNet.