{"Package":"torch","Type":"Package","Title":"Tensors and Neural Networks with 'GPU' Acceleration","Version":"0.0.2","Authors@R":"c(\nperson(\"Daniel\", \"Falbel\", email = \"daniel@rstudio.com\", role = c(\"aut\", \"cre\", \"cph\")),\nperson(\"Javier\", \"Luraschi\", email = \"javier@rstudio.com\", role = c(\"aut\", \"cph\")),\nperson(\"Dmitriy\", \"Selivanov\", role = c(\"ctb\")),\nperson(\"Athos\", \"Damiani\", role = c(\"ctb\")),\nperson(family = \"RStudio\", role = c(\"cph\"))\n)","Description":"Provides functionality to define and train neural networks similar to\n'PyTorch' by Paszke et al (2019) but written entirely in R\nusing the 'libtorch' library. Also supports low-level tensor operations and\n'GPU' acceleration.","License":"MIT + file LICENSE","URL":"https://torch.mlverse.org/docs, https://github.com/mlverse/torch","BugReports":"https://github.com/mlverse/torch/issues","Encoding":"UTF-8","LazyData":"true","SystemRequirements":"C++11, LibTorch (https://pytorch.org/)","LinkingTo":"Rcpp","Imports":"Rcpp, R6, withr, rlang, methods, utils, stats, bit64","RoxygenNote":"7.1.1","Suggests":"testthat (>= 2.1.0), covr, knitr, rmarkdown, magrittr, glue","VignetteBuilder":"knitr","Collate":"'R7.R' 'RcppExports.R' 'tensor.R' 'autograd.R' 'backends.R'\n'codegen-utils.R' 'conditions.R' 'creation-ops.R' 'cuda.R'\n'device.R' 'dimname_list.R' 'dtype.R' 'gen-method.R'\n'gen-namespace-docs.R' 'gen-namespace-examples.R'\n'gen-namespace.R' 'generator.R' 'help.R' 'indexing.R'\n'install.R' 'lantern_load.R' 'lantern_sync.R' 'layout.R'\n'memory_format.R' 'utils-data.R' 'nn.R' 'nn-activation.R'\n'nn-batchnorm.R' 'nn-conv.R' 'nn-dropout.R' 'nn-init.R'\n'nn-linear.R' 'nn-loss.R' 'nn-pooling.R' 'nn-rnn.R'\n'nn-sparse.R' 'nn-utils-rnn.R' 'nn-utils.R' 'nn_adaptive.R'\n'nnf-activation.R' 'nnf-batchnorm.R' 'nnf-conv.R'\n'nnf-distance.R' 'nnf-dropout.R' 'nnf-embedding.R' 'nnf-fold.R'\n'nnf-instancenorm.R' 'nnf-linear.R' 'nnf-loss.R'\n'nnf-normalization.R' 'nnf-padding.R' 'nnf-pixelshuffle.R'\n'nnf-pooling.R' 'nnf-upsampling.R' 'nnf-vision.R' 'operators.R'\n'optim.R' 'optim-adam.R' 'optim-sgd.R' 'package.R' 'qscheme.R'\n'quantization.R' 'reduction.R' 'save.R' 'scalar.R' 'storage.R'\n'tensor_list.R' 'tensor_options.R' 'type-info.R'\n'utils-data-collate.R' 'utils-data-dataloader.R'\n'utils-data-enum.R' 'utils-data-fetcher.R'\n'utils-data-sampler.R' 'utils.R' 'variable_list.R'\n'with-indices.R' 'wrapers.R'","NeedsCompilation":"yes","Packaged":"2020-08-29 15:28:15 UTC; dfalbel","Author":"Daniel Falbel [aut, cre, cph],\nJavier Luraschi [aut, cph],\nDmitriy Selivanov [ctb],\nAthos Damiani [ctb],\nRStudio [cph]","Maintainer":"Daniel Falbel ","Repository":"CRAN","Date/Publication":"2020-08-31 07:00:02 UTC","repoType":"cran","tarballUrl":"ftp://cran.r-project.org/pub/R/src/contrib/torch_0.0.2.tar.gz","jsonAuthors":[{"name":"Daniel Falbel","email":"daniel@rstudio.com","maintainer":true},{"name":"Javier Luraschi","email":"javier@rstudio.com","maintainer":false},{"name":"Dmitriy Selivanov","email":null,"maintainer":false},{"name":"Athos Damiani","email":null,"maintainer":false},{"name":" RStudio","email":null,"maintainer":false}],"readme":"\n\n\ntorch

\n============================================================================================================\n\n[![Lifecycle:\nexperimental](https://img.shields.io/badge/lifecycle-experimental-orange.svg)](https://www.tidyverse.org/lifecycle/#experimental)\n![R build\nstatus](https://github.com/mlverse/torch/workflows/Test/badge.svg)\n[![CRAN\nstatus](https://www.r-pkg.org/badges/version/torch)](https://CRAN.R-project.org/package=torch)\n[![](https://cranlogs.r-pkg.org/badges/torch)](https://cran.r-project.org/package=torch)\n\nInstallation\n------------\n\nRun:\n\n remotes::install_github(\"mlverse/torch\")\n\nAt the first package load additional software will be installed.\n\nExample\n-------\n\nCurrently this package is only a proof of concept and you can only\ncreate a Torch Tensor from an R object. And then convert back from a\ntorch Tensor to an R object.\n\n library(torch)\n x <- array(runif(8), dim = c(2, 2, 2))\n y <- torch_tensor(x, dtype = torch_float64())\n y\n #> torch_tensor \n #> (1,.,.) = \n #> 0.5406 0.8648\n #> 0.3097 0.9715\n #> \n #> (2,.,.) = \n #> 0.1309 0.8992\n #> 0.4849 0.1902\n #> [ CPUDoubleType{2,2,2} ]\n identical(x, as_array(y))\n #> [1] TRUE\n\n### Simple Autograd Example\n\nIn the following snippet we let torch, using the autograd feature,\ncalculate the derivatives:\n\n x <- torch_tensor(1, requires_grad = TRUE)\n w <- torch_tensor(2, requires_grad = TRUE)\n b <- torch_tensor(3, requires_grad = TRUE)\n y <- w * x + b\n y$backward()\n x$grad\n #> torch_tensor \n #> 2\n #> [ CPUFloatType{1} ]\n w$grad\n #> torch_tensor \n #> 1\n #> [ CPUFloatType{1} ]\n b$grad\n #> torch_tensor \n #> 1\n #> [ CPUFloatType{1} ]\n\n### Linear Regression\n\nIn the following example we are going to fit a linear regression from\nscratch using torch’s Autograd.\n\n**Note** all methods that end with `_` (eg. `sub_`), will modify the\ntensors in place.\n\n x <- torch_randn(100, 2)\n y <- 0.1 + 0.5*x[,1] - 0.7*x[,2]\n\n w <- torch_randn(2, 1, requires_grad = TRUE)\n b <- torch_zeros(1, requires_grad = TRUE)\n\n lr <- 0.5\n for (i in 1:100) {\n y_hat <- torch_mm(x, w) + b\n loss <- torch_mean((y - y_hat$squeeze(1))^2)\n \n loss$backward()\n \n with_no_grad({\n w$sub_(w$grad*lr)\n b$sub_(b$grad*lr) \n \n w$grad$zero_()\n b$grad$zero_()\n })\n }\n print(w)\n #> torch_tensor \n #> 1e-09 *\n #> 5.2672\n #> -6.7969\n #> [ CPUFloatType{2,1} ]\n print(b) \n #> torch_tensor \n #> 0.01 *\n #> -9.6802\n #> [ CPUFloatType{1} ]\n\nContributing\n------------\n\nNo matter your current skills it’s possible to contribute to `torch`\ndevelopment. See the contributing guide for more information.","jobInfo":{"package":"torch","version":"0.0.2","parsingStatus":"success","parserVersion":1,"parsedAt":"2020-08-31T09:33:11+0000"}}

Description

Usage

Arguments