{"id":18015873,"url":"https://github.com/reinterpretcat/mikrograd","last_synced_at":"2025-04-04T15:22:49.064Z","repository":{"id":69451151,"uuid":"566335058","full_name":"reinterpretcat/mikrograd","owner":"reinterpretcat","description":"A toy neural networks library with zero* dependencies","archived":false,"fork":false,"pushed_at":"2022-11-17T19:17:58.000Z","size":68,"stargazers_count":2,"open_issues_count":0,"forks_count":0,"subscribers_count":2,"default_branch":"master","last_synced_at":"2025-02-10T01:14:45.362Z","etag":null,"topics":["autograd","micrograd","neural-networks","rust"],"latest_commit_sha":null,"homepage":"","language":"Rust","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/reinterpretcat.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2022-11-15T13:09:58.000Z","updated_at":"2023-04-06T21:26:36.000Z","dependencies_parsed_at":"2023-02-22T19:16:16.709Z","dependency_job_id":null,"html_url":"https://github.com/reinterpretcat/mikrograd","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/reinterpretcat%2Fmikrograd","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/reinterpretcat%2Fmikrograd/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/reinterpretcat%2Fmikrograd/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/reinterpretcat%2Fmikrograd/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/reinterpretcat","download_url":"https://codeload.github.com/reinterpretcat/mikrograd/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":247198807,"owners_count":20900150,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["autograd","micrograd","neural-networks","rust"],"created_at":"2024-10-30T04:15:17.982Z","updated_at":"2025-04-04T15:22:49.037Z","avatar_url":"https://github.com/reinterpretcat.png","language":"Rust","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Description\n\nA toy example of a tiny Autograd engine inspired by https://github.com/karpathy/micrograd\n\nImplements backpropagation (reverse-mode autodiff) over a dynamically built DAG and a small neural networks library\non top of it with a PyTorch-like API.\n\n# Details\n\n* almost zero dependencies (only `rand` crate to initialize weights with uniform distribution)\n* use Rc\u003cRefCell\u003c\u003e\u003e to share mutable state (gradients and weights)\n\n# Usage\n\nThe full example see here: [examples/moons.rs](examples/moons.rs). Some key aspects:\n\nCreate model and test data:\n\n```rust\n    // create model\n    let mut model = mikrograd::new_mlp(2, \u0026[16, 16, 1]);\n\n    // generate test data\n    let (x_data, y_labels) = make_moons(n_samples);\n```\n\nDefine loss function:\n```rust\nfn loss(x_data: \u0026Array\u003cf64, Ix2\u003e, y_labels: \u0026Array\u003cf64, Ix1\u003e, model: \u0026MLP) -\u003e (Value, f64) {\n    let inputs = x_data.map_axis(Axis(1), |data| data.mapv(mikrograd::new_value));\n\n    // forward the model to get scores\n    let scores = inputs.mapv(|input| model.call(input.as_slice().unwrap())[0].clone());\n\n    //svm \"max-margin\" loss\n    let losses = ndarray::Zip::from(y_labels).and(\u0026scores).map_collect(|\u0026yi, scorei| (1. + -yi * scorei).relu());\n    let losses_len = losses.len() as f64;\n    let data_loss = losses.into_iter().sum::\u003cValue\u003e() / losses_len;\n\n    // L2 regularization\n    let alpha = 1E-4;\n    let reg_loss = alpha * model.parameters().map(|p| p * p).sum::\u003cValue\u003e();\n    let total_loss = data_loss + reg_loss;\n\n    // also get accuracy\n    let accuracy =\n        ndarray::Zip::from(y_labels).and(\u0026scores).map_collect(|\u0026yi, scorei| (yi \u003e 0.) == (scorei.get_data() \u003e 0.));\n    let accuracy = accuracy.fold(0., |acc, \u0026hit| acc + if hit { 1. } else { 0. }) / accuracy.len() as f64;\n\n    return (total_loss, accuracy);\n}\n```\n\nRun optimization loop:\n\n```rust\n    for k in 0..100 {\n        // forward\n        let (total_loss, accuracy) = loss(\u0026x_data, \u0026y_labels, \u0026model);\n\n        // backward\n        model.zero_grad();\n        total_loss.backward();\n\n        // update (sgd)\n        let learning_rate = 1. - 0.9 * k as f64 / 100.;\n        for p in model.parameters_mut() {\n            p.set_data(p.get_data() - learning_rate * p.get_grad());\n        }\n\n        println!(\"step {} loss {}, accuracy {:.2}%\", k, total_loss.get_data(), accuracy * 100.);\n    }\n```\n\n\"Poor-man's\" visualization of decision boundary:\n\n![moons](moons_100.png)","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Freinterpretcat%2Fmikrograd","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Freinterpretcat%2Fmikrograd","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Freinterpretcat%2Fmikrograd/lists"}