See the version list below for details.
Install-Package Proxem.NumNet -Version 1.5.0
dotnet add package Proxem.NumNet --version 1.5.0
<PackageReference Include="Proxem.NumNet" Version="1.5.0" />
paket add Proxem.NumNet --version 1.5.0
#r "nuget: Proxem.NumNet, 1.5.0"
// Install Proxem.NumNet as a Cake Addin #addin nuget:?package=Proxem.NumNet&version=1.5.0 // Install Proxem.NumNet as a Cake Tool #tool nuget:?package=Proxem.NumNet&version=1.5.0
NumNet is an optimized library for matrix operations and scientific programming written in C# and developped at Proxem. NumNet is inspired by python's numpy library to facilitate its use by python developpers.
Table of contents
NumNet was developped in .Net Standard 2.0 and is compatible with both .Net Framwork and .Net Core thus working on Windows and Linux platform. For Mac OS users there shouldn't be any problem but we didn't test extensively.
NumNet relies on BlasNet for the low level operations on arrays. See BlasNet documentation for further informations on how to use Intel's MKL for low level operations.
To create an empty 2-dimensional array of dimension 3 and 4 you can use
var zeroArray = NN.Zeros(3, 4);
The follwing creates a 1-dimensional array with all even number from 0 to 40
var range = NN.Range(0, 40, step: 2);
To reshape the previous array to a 2-d dimension array use
var 2dRange = range.Reshape(4, 5);.
This operation will be a O(1) if possible but it might need to copy the values
(if the initial matrix is transposed or more generaly if the data in the initial array are not contiguous.)
Random initializations are also supported, here are a few examples of the supported distributions
var bern = NN.Random.Bernouilli(0.5, 2, 3); // 2 x 3 matrix var norm = NN.Random.Normal(0, 1, 10, 10); // 10 x 10 normally distributed matrix var unif = NN.Random.Uniform(-1, 1, 5, 6); // 5 x 6 uniform matrix between -1 and 1
Let's start with a 2-d array of size (5 x 6)
var M = NN.Range(30).Reshape(5, 6);
To access a single value in the array we will use
NumNet also supports more complex slicing functions.
To select the first column of the array we will use
var vector = M[Slicer._, 0]; // 'Slicer._' correspond to ':' in numpy
More control on the slices is made using the following
var v0 = M; // [0, 1, 2, 3, 4, 5] var v1 = M[Slicer.Range(0, 3), Slicer.Until(2)]; // [[0, 1],[6, 7],[12, 13]] var v2 = M[1, Slicer.From(1)]; // [7, 8, 9, 10, 11] var v3 = M[Slicer.Range(3, -1), -2]; // [16, 22]
The synthax for operations between multi-dimensional arrays is mostly the same as numpy (with Pascal Case). For instance, matrix multiplications will be done with the follwing code
var M = NN.Random.Normal(0, 1, 3, 4); var N = NN.Random.Bernouilli(0.6, 4, 5); var MN = NN.Dot(M, N) // gives a (3 x 5) matrix var MMNTranspose = NN.Dot(MN.T, M) // gives a (5 x 4) matrix
MN.T stands for the transpose of
The code of NumNet is open-source, you can find it on our github.
If you can't make NumNet work on your computer or if you have any tracks of improvement drop us an e-mail at one of the following address:
NumNet is Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to you under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.
- Proxem.BlasNet (>= 1.5.0)
NuGet packages (2)
Showing the top 2 NuGet packages that depend on Proxem.NumNet:
Package containing a word2vec object for fast nearest neighbors search. Saving and loading format are compatible with python's gensim module.
Library for optimized Deep Learning models. The library is based on python's theano library and offers great tools to create deep learning models. Automatic gradient differentiation is available.
This package is not used by any popular GitHub repositories.