Our AI and Neural Network tools for PyTorch: clean, aligned, readable views of tensors, single neuron models, and multi-layer networks, in Jupyter Notebooks and the console. Replace print(tensor) with show_tensor() and see a tensor's Name, Shape, Datatype, Device, and Values in one aligned block.
Every model knows its kind, the arguments it was built with, its data by phase (train, val, and test) with feature names, classes, and scaling, and its history: how it was created, given its data, trained, and tested. So nearly every function takes only the model and a phase by name.
See any model as its Networks, Layers, and Neurons: each neuron's weights largest first, a mean |w| bar per Layer, activations by class, and which way every feature pushes the decision. No more guessing what a network has learned.
Trace one sample through every Layer of a model, chart each term of a neuron's score, and browse samples one by one with a Jupyter slider. 2D scores, 3D decision planes, and 3D latent spaces spin with stepped sliders.
Split DataFrames and datasets into train, val, and test, train on the CPU or a CUDA GPU while keeping the best validation weights, then chart training curves, ROC curves, confusion matrices, regression fits, largest errors, and models side by side. Classifiers, regressors, convolutional networks, and autoencoders.
Like our NPM packages, MicroCODE Neural follows our MicroCODE coding style: consistent, explicit, simple, and readable. Built on PyTorch, NumPy, pandas, and Matplotlib for Python 3.12+, tested with pytest, and used every day in our own Notebooks. Install it with one command:
pip install mcode-neural
