mateuszzwierzycki/Owl

The Owl.Accord.GH.Common library now includes support for the Accord.NET ActivationNetwork. A new GH\_ActivationNetwork class wraps the Accord Neuro library's ActivationNetwork, enabling serialization, deserialization, and casting within Grasshopper. A corresponding Param\_ActivationNetwork class allows users to pass and store ActivationNetwork objects in Grasshopper definitions, with file-based loading and icon support.

Owl.Accord.GH.Common · high confidence

The Owl.Accord.Extensions library now includes new components for working with the Accord.NET framework and video input. This adds a custom TFSigmoid activation function, extension methods for reconstructing, trimming, duplicating, and evaluating ActivationNetwork instances, and a FrameCapture class for listing and capturing frames from video devices.

Owl.Accord.Extensions · high confidence

The Owl.GH.Networking module now includes new components for MQTT communication. Users can connect to an MQTT broker via the new 'MQTT Connect' component, which opens a form to configure host, port, credentials, and topics. The 'MQTT Publish' component allows sending messages to a specified topic with configurable QoS and retain flags. Additionally, the 'MQTT Read' component processes incoming messages from a buffer, exposing topics and message content as outputs. These components enable real-time IoT data exchange within Grasshopper.

Owl.GH.Networking · high confidence

Added a new Grasshopper component, OwlParamTester, which demonstrates how to use Owl.GH.Common. The component accepts an OwlTensorSet input and outputs the average of the tensor values. The project also includes standard Visual Studio project files, resource files, and an app.config that redirects RhinoCommon to version 5.1.30000.16.

Owl.ParamSetup/Owl.ParamSetup · high confidence

A new QAgent class has been added to the Owl.Learning/QLearning namespace, providing a complete implementation of the Q-Learning algorithm. The class manages a Q-matrix (Tensor) and exposes properties for the discount factor (Gamma) and learning rate (Alpha). It includes methods for choosing actions based on current state and epsilon-greedy exploration, updating Q-values with rewards, and managing random number generation for reproducibility.

Owl.Learning/QLearning · high confidence

The Owl.Learning.Neuro library now includes a complete set of classes to define and train a multi-layer neural network. This includes ActivationNetwork, ActivationLayer, and ActivationNeuron to construct the network architecture, along with BackPropagationLearning to handle the training process. Users can now create networks with configurable layers and neurons, and train them using backpropagation with adjustable learning rate and momentum parameters.

Owl.Learning.Neuro · high confidence

The Owl.Learning project now includes its core assembly metadata and configuration files, including AssemblyInfo.vb (defining version 2.1.0.0 and copyright details), Application.myapp (configuring application behavior), and Settings files (enabling application settings persistence). These changes establish the foundational identity and configuration for the learning module.

Owl.Learning/My Project · high confidence

The repository now includes the complete Owl framework, a set of libraries for machine-learning-oriented data pre- and post-processing, normalization, serialization, and visualization. The release introduces the core Owl.Core library, Owl.Learning for machine learning methods, and various Grasshopper (GH) plugins and extensions (Owl.GH, Owl.Accord.GH, etc.) that enable developers to use Owl types within GH environments. A .gitignore file is also added to manage build artifacts and user-specific files.

(repo-wide) · high confidence

The K-Means clustering algorithm has been enhanced to allow users to provide their own distance metric and cluster-averaging logic. A new \KMeansEngineMetric\ class accepts delegates for these functions, enabling flexible, user-defined clustering behavior instead of relying solely on the default Euclidean-style calculations.

Owl.Learning/Clustering · high confidence

The Owl.GH.Common library has been introduced, providing a suite of base classes and data types for the Grasshopper environment. This includes component base classes like ImageComponentBase for handling image-based UI elements, and OwlComponentBase which establishes a consistent 'Owl' category for new tools. The update also adds Goo wrappers for core data structures—such as Tensor, TensorSet, Network, QAgent, and QLearning—enabling seamless data flow between Grasshopper and the underlying C\# logic. Additionally, a Trigger class and its corresponding Goo wrapper are added to support event-driven workflows within the canvas.

Owl.GH.Common · high confidence

The Owl.Learning module now includes a comprehensive set of new components for machine learning workflows. For supervised learning, users can construct, deconstruct, and compute outputs from neural networks (ConstructNetwork, ConstructNetwork\_Direct, DeconstructNetwork, Compute). Unsupervised learning gains K-Means clustering (KMeansClustering, KMeansClusteringEx), line clustering (ClusterDirections), and Markov chain sequence generation (MarkovChain). Reinforcement learning is supported via components to construct and deconstruct Q-Agents, update Q-values, choose actions, and build Q-matrices (ConstructQLearn, DeconstructQLearn, UpdateQ, ChooseAction, ConstructQMatrix, Matrix2QMatrix). Additionally, a new LifeSaver component provides a graphical interface for managing learning tasks within Grasshopper.

Owl.GH · high confidence

Added new modules for image processing: \ImageConverters\ provides functions to convert between Tensors and Bitmaps (including grayscale and multi-channel formats), while \ImageFunctions\ introduces tensor operations for cropping, mirroring, rotating, and transposing images. Additionally, the \mnistImageReader\ and \mnistLabelReader\ classes were removed, and the \ColorHSL\ structure was moved to the \Structures\ namespace.

Owl.Core · high confidence

Adds a suite of new components for building, training, and visualizing neural networks. The 'Backpropagation' and 'Resilient' components enable supervised training using the Accord.NET library, with options for learning rate, momentum, and iterations. A 'Compute' component allows for forward passes to generate outputs, while 'EvaluateNetwork' calculates error metrics. Network management is supported via 'AssignNetworkValues' and 'GetNetworkValues' to modify or inspect weights and biases. Visualization is enhanced with 'NetPreview' and 'NetworkPreview\_OBSOLETE' for 2D network diagrams, and 'DisplayCompute' for tensor visualization. Additionally, 'TSNE' provides dimensionality reduction, and 'CameraView' enables webcam frame capture.

Owl.Accord.GH · high confidence

The Owl.Learning/Networks area introduces new foundational classes for building neural networks. A new NeuronFunctions namespace provides base and concrete activation functions (Tanh, Linear, ReLU, Sigmoid) with derivative support. An Initializers namespace adds a RandomInitializer for weight/bias initialization. The core Network class now supports mixed activation functions per layer and includes optimized compute methods, enabling more flexible network architectures.

Owl.Learning/Networks · high confidence

The Owl.ParamSetup project has been reorganized to support both 32-bit and 64-bit build configurations, allowing users to choose the appropriate architecture for their environment. The solution file has been renamed and updated to reflect the new project structure, with separate build configurations for Debug32 and Debug64 platforms. This change enables better compatibility with different Grasshopper/Rhino environments by providing targeted builds for each architecture.

Owl.ParamSetup · medium confidence

Moved image processing classes (Convolution, ImageModifier, ImageModifierLayer, ImageNetwork, Normalizer, Pooling, Rectification, Subsampling) from Owl.Core to Owl.Learning/Images. Updated the Convolution class to use Double instead of Integer for kernel values, changed the default for multithreading in ImageModifier.Apply to False, and updated the Normalizer to scale values to a 0-1 range instead of 0-255. Also moved MarkovChain to Owl.Learning/Markov.

Owl.Learning/Images, Owl.Learning/Markov · medium confidence

The build system was reorganized with the addition of several new project files, including Owl.Accord.Extensions, Owl.Accord.GH.Common, Owl.Accord.GH, Owl.GH.Common, Owl.GH.Networking, Owl.GH, Owl.Learning.Neuro, Owl.Learning, and Owl.ParamSetup. These projects now reference specific versions of the Accord (3.8.0), Grasshopper (7.16.22067.13001), and RhinoCommon (7.16.22067.13001) libraries, alongside other dependencies like System.Reactive (5.0.0) and System.Net.Mqtt (0.5.42-beta).

(dependencies) · high confidence

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• CAI 48 → 47 (-0.9)
• Rubric changed (rubric-2026.08.15 → rubric-2026.09.16) — scores are not directly comparable.

• Code Health 92 → 92 (-0.2)
• Architecture 98 → 98 (+0.2)
• Maturity 46 → 42 (-3.5)
• Readiness 23 → 23 (+0.0)
• Security 100 → 100 (+0.0)

• Bounded contexts not declared
• Build did not complete in the analyzer
• Duplicated block (10 lines × 2) (Owl.Accord.GH/Components/Display/NetworkPreview_Attributes.vb)
• Duplicated block (10 lines × 2) (Owl.Core/Images/Image operations/ImageConverters.vb)
• Duplicated block (11 lines × 2) (Owl.GH/Params/Param_OwlFiles.vb)
• Duplicated block (12 lines × 2) (Owl.Learning/Clustering/KMeansEngine.vb)
• Duplicated block (13 lines × 2) (Owl.Learning.Neuro/BackpropagationLearning.cs)
• Duplicated block (15 lines × 2) (Owl.Accord.GH/Components/Display/NetworkPreview_Attributes.vb)
• Duplicated block (5 lines × 3) (Owl.Accord.GH/Components/Backpropagation/BackOwl.vb)
• Duplicated block (5 lines × 5) (Owl.Accord.GH/Components/Backpropagation/BackpropLearning.vb)
• Duplicated block (6 lines × 2) (Owl.GH/Params/Param_OwlFiles.vb)
• Duplicated block (8 lines × 2) (Owl.Accord.GH/Components/Display/NetworkPreview_Attributes.vb)
• Duplicated block (9 lines × 2) (Owl.Accord.GH/Components/Backpropagation/BackpropLearning.vb)
• Duplicated block (9 lines × 2) (Owl.Learning/Clustering/KMeansEngine.vb)
• Duplicated block (9 lines × 2) (Owl.Learning/Clustering/KMeansEngine.vb)
• Further orphaned files (smaller)
• Inconsistency in naming the collection of components: 'Neurons' for a layer, 'Layers' for a network. While distinct, the pattern of pluralizing the component name is consistent. However, there is a parameter naming inconsistency: 'input' vs 'inputs' vs 'desiredOutput' vs 'output'. Specifically, 'input' (singular) is used in BackPropagationLearning.CalculateError and ActivationNeuron.Compute, while 'inputs' (plural) is used in BackPropagationLearning.Run and ActivationLayer.Compute. This is a minor inconsistency in parameter naming for the same concept (the input data).
• Low cohesion: Tensor (LCOM4 5) (Owl.Core/Tensors/Tensor.vb)
• Monorepo: only 1 of 2 solutions was scored
• No exposed public API
• …and 23 more

• Build failed
• Dormant codebase
• Duplicated block (10 lines × 2) (Owl.Accord.GH/Components/Backpropagation/BackpropLearning.vb)
• Duplicated block (11 lines × 2) (Owl.Learning/Clustering/KMeansEngine.vb)
• Duplicated block (11 lines × 2) (Owl.Learning/Clustering/KMeansEngine.vb)
• Duplicated block (12 lines × 2) (Owl.Core/IO/TensorSerialization.vb)
• Duplicated block (12 lines × 3) (Owl.Accord.Extensions/Visualization/Visualization.vb)
• Duplicated block (13 lines × 2) (Owl.GH/Params/Param_OwlFiles.vb)
• Duplicated block (13 lines × 2) (Owl.Learning.Neuro/BackpropagationLearning.cs)
• Duplicated block (16 lines × 2) (Owl.Accord.GH/Components/Display/NetworkPreview_Attributes.vb)
• Duplicated block (17 lines × 2) (Owl.Accord.GH/Components/Display/NetworkPreview_Attributes.vb)
• Duplicated block (19 lines × 2) (Owl.Learning/Clustering/KMeansEngine.vb)
• Duplicated block (20 lines × 2) (Owl.Core/Images/Image operations/ImageConverters.vb)
• Duplicated block (26 lines × 2) (Owl.Core/Images/Image operations/ImageConverters.vb)
• Duplicated block (33–34 lines × 2) (Owl.Accord.GH/Components/Display/NetworkPreview_Attributes.vb)
• Duplicated block (5 lines × 2) (Owl.Accord.GH/Components/Unsupervised/TSNE.vb)
• Duplicated block (5 lines × 3) (Owl.Accord.GH/Components/Backpropagation/BackpropLearning.vb)
• Duplicated block (6 lines × 2) (Owl.Accord.GH/Components/Backpropagation/BackpropLearning.vb)
• Duplicated block (6 lines × 2) (Owl.GH/Params/Param_OwlFiles.vb)
• Duplicated block (6 lines × 3) (Owl.Accord.GH/Components/Backpropagation/BackOwl.vb)
• …and 8 more

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

mateuszzwierzycki/Owl was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.