{"Package":"mlflow","Type":"Package","Title":"Interface to 'MLflow'","Version":"0.7.0","Authors@R":"c(\nperson(\"Matei\", \"Zaharia\", email = \"matei@databricks.com\", role = c(\"aut\", \"cre\")),\nperson(\"Javier\", \"Luraschi\", email = \"javier@rstudio.com\", role = c(\"aut\")),\nperson(\"Kevin\", \"Kuo\", role = c(\"aut\"), email = \"kevin.kuo@rstudio.com\",\ncomment = c(ORCID = \"0000-0001-7803-7901\")),\nperson(family = \"RStudio\", role = c(\"cph\"))\n)","Maintainer":"Matei Zaharia ","Description":"R interface to 'MLflow', open source platform for the complete machine\nlearning life cycle, see . This package supports installing\n'MLflow', tracking experiments, creating and running projects, and saving and\nserving models.","License":"Apache License 2.0","URL":"https://github.com/mlflow/mlflow","BugReports":"https://github.com/mlflow/mlflow/issues","SystemRequirements":"MLflow (https://www.mlflow.org/)","Encoding":"UTF-8","LazyData":"true","Depends":"R (>= 3.1.2)","Imports":"aws.s3, forge, fs, git2r, httpuv, httr, jsonlite, openssl,\nprocessx, reticulate, packrat, purrr, swagger, withr, xml2,\nyaml, pryr, rlang (>= 0.2.0)","RoxygenNote":"6.1.0","Suggests":"covr, keras, lintr, testthat","NeedsCompilation":"no","Packaged":"2018-09-28 19:17:25 UTC; tomas","Author":"Matei Zaharia [aut, cre],\nJavier Luraschi [aut],\nKevin Kuo [aut] (),\nRStudio [cph]","Repository":"CRAN","Date/Publication":"2018-10-06 21:30:03 UTC","repoType":"cran","tarballUrl":"ftp://cran.r-project.org/pub/R/src/contrib/mlflow_0.7.0.tar.gz","jsonAuthors":[{"name":"Matei Zaharia","email":"matei@databricks.com","maintainer":true},{"name":"Javier Luraschi","email":"javier@rstudio.com","maintainer":false},{"name":"Kevin Kuo","email":"kevin.kuo@rstudio.com","maintainer":false},{"name":" RStudio","email":null,"maintainer":false}],"readme":"mlflow: R interface for MLflow\n================\n\n[![Build\nStatus](https://travis-ci.org/rstudio/mlflow.svg?branch=master)](https://travis-ci.org/rstudio/mlflow)\n[![CRAN\\_Status\\_Badge](https://www.r-pkg.org/badges/version/mlflow)](https://cran.r-project.org/package=mlflow)\n[![codecov](https://codecov.io/gh/mlflow/mlflow/branch/master/graph/badge.svg)](https://codecov.io/gh/mlflow/mlflow)\n\n - Install [MLflow](https://mlflow.org/) from R to track experiments\n locally.\n - Connect to MLflow servers to share experiments with others.\n - Use MLflow to export models that can be served locally and remotely.\n\n## Installation\n\nInstall `mlflow` from CRAN followed by installing the `mlflow` runtime\nas follows:\n\n``` r\ninstall.packages(\"mlflow\")\nmlflow::mlflow_install()\n```\n\nNotice also that [Anaconda](https://www.anaconda.com/download/) or\n[Miniconda](https://conda.io/miniconda.html) need to be manually\ninstalled.\n\n### Developemnt\n\nInstall the `mlflow` package as follows:\n\n``` r\ndevtools::install_github(\"mlflow/mlflow\", subdir = \"mlflow/R/mlflow\")\n```\n\nThen install the latest released `mlflow` runtime.\n\n``` r\n# Install latest released version\nmlflow::mlflow_install()\n```\n\nHowever, currently, the development runtime of `mlflow` is also\nrequired; which means you also need to download or clone the `mlflow`\nGitHub repo:\n\n``` bash\ngit clone https://github.com/mlflow/mlflow\n```\n\nAnd upgrade the runtime to the development version as follows:\n\n``` r\n# Upgrade to the latest development version\nreticulate::conda_install(\"r-mlflow\", \"\", pip = TRUE)\n```\n\n## Tracking\n\nMLflow Tracking allows you to logging parameters, code versions,\nmetrics, and output files when running R code and for later visualizing\nthe results.\n\nMLflow allows you to group runs under experiments, which can be useful\nfor comparing runs intended to tackle a particular task. You can create\nand activate a new experiment locally using `mlflow` as follows:\n\n``` r\nlibrary(mlflow)\nmlflow_set_experiment(\"Test\")\n```\n\nThen you can list view your experiments from MLflows user interface by\nrunning:\n\n``` r\nmlflow_ui()\n```\n\n

\n\nYou can also use a MLflow server to track and share experiments, see\n[running a tracking\nserver](https://www.mlflow.org/docs/latest/tracking.html#running-a-tracking-server),\nand then make use of this server by running:\n\n``` r\nmlflow_set_tracking_uri(\"http://tracking-server:5000\")\n```\n\nOnce the tracking url is defined, the experiments will be stored and\ntracked in the specified server which others will also be able to\naccess.\n\n## Projects\n\nAn MLflow Project is a format for packaging data science code in a\nreusable and reproducible way.\n\nMLflow projects can be [explicitly\ncreated](https://www.mlflow.org/docs/latest/projects.html#specifying-projects)\nor implicitly used by running `R` with `mlflow` from the terminal as\nfollows:\n\n``` bash\nmlflow run examples/r_wine --entry-point train.R\n```\n\nNotice that is equivalent to running from `examples/r_wine`,\n\n``` bash\nRscript -e \"mlflow::mlflow_source('train.R')\"\n```\n\nand `train.R` performing training and logging as follows:\n\n``` r\nlibrary(mlflow)\n\n# read parameters\ncolumn <- mlflow_log_param(\"column\", 1)\n\n# log total rows\nmlflow_log_metric(\"rows\", nrow(iris))\n\n# train model\nmodel <- lm(\n Sepal.Width ~ x,\n data.frame(Sepal.Width = iris$Sepal.Width, x = iris[,column])\n)\n\n# log models intercept\nmlflow_log_metric(\"intercept\", model$coefficients[[\"(Intercept)\"]])\n```\n\n### Parameters\n\nYou will often want to parameterize your scripts to support running and\ntracking multiple experiments. Ypu can define parameters with type under\na `params_example.R` example as follows:\n\n``` r\nlibrary(mlflow)\n\n# define parameters\nmy_int <- mlflow_param(\"my_int\", 1, \"integer\")\nmy_num <- mlflow_param(\"my_num\", 1.0, \"numeric\")\n\n# log parameters\nmlflow_log_param(\"param_int\", my_int)\nmlflow_log_param(\"param_num\", my_num)\n```\n\nThen run `mlflow run` with custom parameters as\n follows\n\n mlflow run tests/testthat/examples/ --entry-point params_example.R -P my_int=10 -P my_num=20.0 -P my_str=XYZ\n\n === Created directory /var/folders/ks/wm_bx4cn70s6h0r5vgqpsldm0000gn/T/tmpi6d2_wzf for downloading remote URIs passed to arguments of type 'path' ===\n === Running command 'source /miniconda2/bin/activate mlflow-da39a3ee5e6b4b0d3255bfef95601890afd80709 && Rscript -e \"mlflow::mlflow_source('params_example.R')\" --args --my_int 10 --my_num 20.0 --my_str XYZ' in run with ID '191b489b2355450a8c3cc9bf96cb1aa3' === \n === Run (ID '191b489b2355450a8c3cc9bf96cb1aa3') succeeded ===\n\nRun results that we can view with `mlflow_ui()`.\n\n## Models\n\nAn MLflow Model is a standard format for packaging machine learning\nmodels that can be used in a variety of downstream tools—for example,\nreal-time serving through a REST API or batch inference on Apache Spark.\nThey provide a convention to save a model in different “flavors” that\ncan be understood by different downstream tools.\n\nTo save a model use `mlflow_save_model()`. For instance, you can add the\nfollowing lines to the previous `train.R` script:\n\n``` r\n# train model (...)\n\n# save model\nmlflow_save_model(\n crate(~ stats::predict(model, .x), model)\n)\n```\n\nAnd trigger a run with that will also save your model as follows:\n\n``` bash\nmlflow run train.R\n```\n\nEach MLflow Model is simply a directory containing arbitrary files,\ntogether with an MLmodel file in the root of the directory that can\ndefine multiple flavors that the model can be viewed in.\n\nThe directory containing the model looks as follows:\n\n``` r\ndir(\"model\")\n```\n\n ## [1] \"crate.bin\" \"MLmodel\"\n\nand the model definition `model/MLmodel` like:\n\n``` r\ncat(paste(readLines(\"model/MLmodel\"), collapse = \"\\n\"))\n```\n\n ## flavors:\n ## crate:\n ## version: 0.1.0\n ## model: crate.bin\n ## time_created: 18-09-27T19:06:55.55.85\n ## run_id: c2e91ac015564ccaa711480c3effd917\n\nLater on, the R model can be deployed which will perform predictions\nusing\n `mlflow_rfunc_predict()`:\n\n``` r\nmlflow_rfunc_predict(\"model\", data = data.frame(x = c(0.3, 0.2)))\n```\n\n ## Warning in mlflow_snapshot_warning(): Running without restoring the\n ## packages snapshot may not reload the model correctly. Consider running\n ## 'mlflow_restore_snapshot()' or setting the 'restore' parameter to 'TRUE'.\n\n ## 3.400381396714573.40656987651099\n\n ## 1 2 \n ## 3.400381 3.406570\n\n## Deployment\n\nMLflow provides tools for deployment on a local machine and several\nproduction environments. You can use these tools to easily apply your\nmodels in a production environment.\n\nYou can serve a model by running,\n\n``` bash\nmlflow rfunc serve model\n```\n\nwhich is equivalent to\nrunning,\n\n``` bash\nRscript -e \"mlflow_rfunc_serve('model')\"\n```\n\n

\n\nYou can also run:\n\n``` bash\nmlflow rfunc predict model data.json\n```\n\nwhich is equivalent to running,\n\n``` bash\nRscript -e \"mlflow_rfunc_predict('model', 'data.json')\"\n```\n\n## Dependencies\n\nWhen running a project, `mlflow_snapshot()` is automatically called to\ngenerate a `r-dependencies.txt` file which contains a list of required\npackages and versions.\n\nHowever, restoring dependencies is not automatic since it’s usually an\nexpensive operation. To restore dependencies run:\n\n``` r\nmlflow_restore_snapshot()\n```\n\nNotice that the `MLFLOW_SNAPSHOT_CACHE` environment variable can be set\nto a cache directory to improve the time required to restore\ndependencies.\n\n## RStudio\n\nTo enable fast iteration while tracking with MLflow improvements over a\nmodel, [RStudio 1.2.897](https://dailies.rstudio.com/) an be configured\nto automatically trigger `mlflow_run()` when sourced. This is enabled by\nincluding a `# !source mlflow::mlflow_run` comment at the top of the R\nscript as\nfollows:\n\n

\n\n## Contributing\n\nSee the [MLflow contribution guidelines](../../../CONTRIBUTING.rst).","jobInfo":{"package":"mlflow","version":"0.7.0","parsingStatus":"success","parserVersion":1,"parsedAt":"2018-11-12T23:29:48+0000"}}

Description

Usage

Arguments

Details

See Also