Skip to main content

Crate mathverse_ai

Crate mathverse_ai 

Source
Expand description

AI/ML mathematical primitives: tensors, activations, losses, optimizers, attention, layers, models.

§Features

  • N-dimensional tensor with broadcasting, reshaping, and element-wise math
  • Activation functions: ReLU, GELU, Swish, Mish, Softmax, Sigmoid, Tanh, and more
  • Loss functions: MSE, MAE, Huber, Cross-Entropy, Binary Cross-Entropy, KL Divergence, Hinge
  • Evaluation metrics: Accuracy, Precision, Recall, F1, Confusion Matrix, ROC AUC, R², Explained Variance
  • Stateful optimizers: SGD (with momentum), Adam, AdamW
  • Learning rate schedulers: Constant, Step Decay, Cosine Annealing, Linear Warmup
  • Attention math: QKV projection, scaled dot-product attention, multi-head attention
  • Positional encodings: sinusoidal (additive), rotary (RoPE)
  • Layer/Batch/RMS normalization

Re-exports§

pub use tensor::Tensor;
pub use activations::relu;
pub use activations::sigmoid;
pub use activations::softmax;
pub use activations::gelu;
pub use activations::swish;
pub use activations::tanh as tanh_act;
pub use losses::mse;
pub use losses::cross_entropy;
pub use losses::binary_cross_entropy_with_logits;
pub use metrics::accuracy;
pub use metrics::f1;
pub use metrics::confusion_matrix;
pub use optimizers::Sgd;
pub use optimizers::Adam;
pub use optimizers::AdamW;
pub use optimizers::LrScheduler;
pub use optimizers::Schedule;
pub use attention::scaled_dot_product_attention;
pub use attention::multi_head_attention;
pub use attention::sinusoidal_encoding;
pub use attention::apply_rope;
pub use layers::Linear;
pub use layers::LayerNorm;
pub use layers::BatchNorm;
pub use layers::Dropout;
pub use models::Sequential;
pub use models::MLP;
pub use models::TransformerBlock;
pub use models::Activation;
pub use data::DataLoader;
pub use data::Batch;
pub use data::train_test_split;

Modules§

activations
Activation functions: element-wise and softmax variants with derivatives.
attention
Attention math: QKV projection, scaled dot-product attention, multi-head attention, sinusoidal positional encoding, rotary embeddings (RoPE).
attention_adv
Advanced attention mechanisms: flash, grouped query, cross, and linear attention.
autograd
Automatic differentiation via reverse-mode autodiff (backpropagation).
conv_transpose
Transposed convolution (deconvolution) for 1D and 2D tensors.
data
Data loading utilities: DataLoader with batching and shuffling.
generative
Generative model components: VAE, GAN, and diffusion utilities.
layers
Neural network layers: Linear, Conv1d, Conv2d, LayerNorm, BatchNorm, Dropout.
losses
Loss functions: regression (MSE, MAE, Huber) and classification (cross-entropy, etc.).
metrics
Evaluation metrics: classification (accuracy, precision, recall, F1, ROC AUC) and regression (MSE, MAE, R², explained variance).
models
Sequential model container: chains layers + activations for forward pass.
optimizers
Stateful optimizers: SGD (with momentum + weight decay), Adam, AdamW, and learning rate schedulers.
recurrent
Recurrent layers: vanilla RNN, LSTM, and GRU forward passes.
registry
Model registry: save/load tensors to text files, parameter counting.
tensor
N-dimensional tensor with row-major layout, broadcasting, and math ops.
vision
Vision Transformer (ViT) components: patch embedding, CLS token, positional embeddings.