GPU-powered Deep Learning for Time Series with modeltime.gluonts on Saturn Cloud

saturn_cloud_modeltime_gluonts_startup_script.sh

# Saturn Cloud Startup Script for Running GPU-enabled modeltime.gluonts
#
# How to use:
#   1. Create an account on saturncloud.io
#   2. Create a new project in Saturn Cloud
#   3. Select image "saturncloud/saturn-gpu:2020.12.23" as image for your project
#   4. Copy and paste code below into "Advanced Settings (optional)" > "Start Script (Bash)".
#      (This code will install R, a Jupyter R kernel, modeltime.gluonts, and prepare a Python environment
#      for executing GluonTS.)
#   5. The startup script takes about 40 min to run. After that, you can start Jupyter Lab in your Saturn Cloud
#      project and run GPU-enabled modeltime.gluonts in a new Jupyter Notebook that uses the "R" kernel.
#
# More info about GPUs in modeltime.gluonts: https://business-science.github.io/modeltime.gluonts/articles/using-gpus.html

sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys E298A3A825C0D65DFD57CBB651716619E084DAB9
sudo su -c "echo 'deb https://cran.rstudio.com/bin/linux/ubuntu bionic-cran40/' >>/etc/apt/sources.list"
sudo apt update
sudo apt -y install r-base libv8-dev libcurl4-openssl-dev libssl-dev
mkdir -p ~/.local/lib/R
export R_LIBS_USER=~/.local/lib/R
R -e "install.packages(c('IRkernel','tidymodels','tidyverse','devtools'))"
R -e "IRkernel::installspec()"
R -e "devtools::install_github('business-science/modeltime.gluonts')"
conda run -n saturn pip install gluonts pandas numpy pathlib mxnet-cu101

try_it_yourself_example.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Example: GPU-powered Deep Learning for Time Series with modeltime.gluonts on Saturn Cloud "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Set GLUONTS_PYTHON environment variable to point modeltime.gluonts to Python environment with mxnet installed\n",
    "Sys.setenv(GLUONTS_PYTHON = '/srv/conda/envs/saturn/bin/python')\n",
    "\n",
    "# Import modeltime.gluonts\n",
    "library(modeltime.gluonts)\n",
    "\n",
    "# Check if modeltime.gluonts can find the the Python environment (should be true)\n",
    "is_gluonts_activated()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_fit_deepar <- deep_ar(\n",
    "  id                    = \"id\",\n",
    "  freq                  = \"M\",\n",
    "  prediction_length     = 24,\n",
    "  lookback_length       = 36,\n",
    "  epochs                = 10, \n",
    "  num_batches_per_epoch = 500,\n",
    "  learn_rate            = 0.001,\n",
    "  num_layers            = 3,\n",
    "  num_cells             = 80,\n",
    "  dropout               = 0.10\n",
    ") %>%\n",
    "  set_engine(\"gluonts_deepar\") %>%\n",
    "  fit(value ~ date + id, m750)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "R",
   "language": "R",
   "name": "ir"
  },
  "language_info": {
   "codemirror_mode": "r",
   "file_extension": ".r",
   "mimetype": "text/x-r-source",
   "name": "R",
   "pygments_lexer": "r",
   "version": "4.0.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}

It turns out SaturnCloud has turned this into an image so we don't need to wait the 40 minutes for an image to build.
514354729342.dkr.ecr.us-east-2.amazonaws.com/saturn-r-modeltime-gpu:2021.03.01.02.45