return an ndarray with NULL data NULL shape, be careful! WARNING**: deprecated

since NdArray has not implemented the template for usize, so we have to return NdArray<f32> instead todo

return the index bases for each dimension by given the shape Example: - shape: \[2, 3, 4\] - return: \[12, 4, 1\] - use: get the data of a index \[0,

numpy with 4 threads, 7.7 seconds rust version 30 seconds~

this operator is for matrix multiplication see **point_multiply** method if you want Hadamard Product

create a dataset with (X, Y) T: the label type; usually usize is for classification task; f32 is for regression task features: input features, X label

convert a dataset to Dataloader for batch iterations E: Dataset or &Dataset for both f32(regression) and usize(classification) shuffle: whether shuffl

Naive Bayes model for classification task class_num: # of class num of a classification dataset is_discrete_feature: you have to tell the model which

define a decision tree (not built yet), f32 for regression task, usize for classification task min_sample_split: it will not try to split the tree if

Support Vector Machine for binary classification only supports linear kernel, using SGD to optimize hinge loss update = lambda * (regularization(w)

linear regression model with MeanSquaredError optimized by mini-batch SGD feature_size: the input size, usually the feature size output_size: the outp

logistic regression model with CrossEntropy optimized by mini-batch SGD for classification task feature_size: the input size class_num: # of classes,

k: k nearest neighbours weighting: weighting the neibours, default is these neighbours are equal features: \[batch, feature\] labels: \[batch\] total_

build K-Means model with unsupervised learning k: number of clusters max_iter: max iterations p: decide the distance, parameter of minkowski distance

create a matrix with shape \[n, m\] + bias \[n\] insize: m outsize: n

create a neural network from the given ModuleList blocks: vector of enum NNModule

from the specify shape

return the nearest cluster idx, note that it is NOT the classification prediction

predict a batch return: vector of T(f32 or usize)

create an NdArray filled with random numbers by the given shape and seed shape: specify the shape seed: default is set to 0

update the weights and bias with the grad_w and grad_b, respectively reduction: len(batch): average over the batch 1: sum the batch lr: learning rate