Skip to content
This repository has been archived by the owner on Mar 14, 2023. It is now read-only.
/ experiment-scripts Public archive

Automatically run experiments, upload results to google docs, analyze them using R, and include the analysis in LaTeX.

Notifications You must be signed in to change notification settings

edmcman/experiment-scripts

Repository files navigation

Overview

This project is a set of skeleton scripts for running experiments, analyzing them, and reporting the results in an academic paper. Specifically, the scripts:

  1. Automatically runs experiments using a script and uploads the results to a google doc and/or rabbitmq queue.

  2. Downloads the results into R

  3. Generates statistics and figures

  4. Incorporates the statistics and figures into LaTeX with Sweave

This process has a number of benefits for academic paper writing:

  • You can watch your experiments progress by opening google docs and watching the new data come in. This can often give you an early indication of the effect, if any, your most recent change will have.

  • You can run experiments up to the deadline. Got last minute improved numbers? Simply run make and regenerate the paper. Did your last run hurt your numbers? Simply revert your google doc to the last good version.

  • Years later, you will always be able to find your experimental data. Need to make a new figure? No problem. Need to share your data? Just share the google doc, and optionally share your R script for analyzing it.

Quick Start

  • Add google docs spreadsheet key and login information to config.py.

  • Run python autoexp_pool.py

  • Edit analyze.R to use your own google docs spreadsheet (you can get this url from File -> Publish to the web).

  • Run make

Details

The scripts in this project run a fictitious experiment that measures how long it takes to type my name (Ed) and the name of my co-author (Thanassis). Thanassis helped me write some of these scripts.

The main experiment script is autoexp.py. It is written in python, and is primarily set up to time external commands. The current version "measures" the time it takes to type a name by generating a random number. However, it should be easily adapted to real experiments. Before it can be used, you must put your google account information and the google docs spreadsheet key in the config.py file. You can find a spreadsheet's key by looking at its url. You can run the experiment using multiple cores on a single machine by running autoexp_pool.py. Alternatively, you can run the experiment on multiple machines with the help of a RabbitMQ server by running autoexp_producer.py on any machine, and then running autoexp_consumer.py on each worker machine. In either case, the scripts should add a new table called paper to the google docs spreadsheet. It should look like this. If you look at the spreadsheet while the script is running, you should be able to see each row being added to the spreadsheet. This is more useful for experiments that take hours or days to run. Alternatively, for experiments that produce outputs too large for a spreadsheet, you can save the results in a RabbitMQ queue by setting use_google=False and output_rabbitmq=True in config.py. You can then download the results into a csv file by running autoexp_download_output.py.

analyze.R is an R script that analyzes the uploaded data. It reads the experiment data directly from google docs, counts the number of samples, computes the mean time to "type" both Ed and Thanassis, and then produces two figures. You will probably want to edit this script so that it uses your own google docs csv file instead of mine. I often start analyzing experimental data by opening R and running source("analyze.R") to download the experimental data. There are many tools inside of R for exploratory data analysis, but I personally prefer visualization using the ggplot2 package.

Finally, the results from analyze.R can also be incorporated into a LaTeX paper. This is done using the Sweave file Stats.Rnw, which creates commands for each statistic in R that needs to be referenced in the paper. An example LaTeX paper is in paper.tex. The final step is to use Makefile to process the Sweave file and run LaTeX. The final result should produce a file similar to paper.pdf.

It is also possible to generate nice figures for inclusion in Powerpoint slides. To produce histsummary-slides.emf, which is suitable for inclusion in Windows Powerpoint, convert histsummary-slides.svg to EMF format using Inkscape on Windows. An example can be seen in slides.pptx. Note that Mac Powerpoint cannot view EMF files properly. For Mac Powerpoint, you can simply use histsummary-slides.pdf, which is generated by analyze.R.

About

Automatically run experiments, upload results to google docs, analyze them using R, and include the analysis in LaTeX.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published