Evidlo / redrum

Reddit wallpaper changer. Does math to rank images.

Redrum - Reddit Wallpaper Downloader and Ranker ==============================================

Redrum is a Reddit wallpaper downloader which scores wallpapers and selects the best based on resolution, aspect ratio, and number of views. It remembers which wallpapers were selected previously so you never see the same image twice.

Install the systemd units to run the script every two hours.

Installation

1. Install through pip

   pip3 install redrum

2. (optional) Install service for automatic wallpaper changing

   # copy service files
   cp -u systemd/* ~/.config/systemd/user/
   
   # enable and start systemd timer
   systemctl --user enable redrum.timer
   systemctl --user start redrum.timer
   
   # the service can be triggered manually as well
   systemctl --user start redrum

   note: If using a python3 virtualenv, change ExecStart in redrum.service to /path/to/venv/bin/redrum

Usage

Redrum creates a config at ~/.config/redrum.ini automatically. You should update this file with your screen resolution and preferred subreddits before executing.

>>> redrum
No previous score cache found at /home/evan/.cache/redrum_cache.json.
Indexing page #0 from subreddit winterporn
Indexing page #1 from subreddit winterporn
Indexing page #2 from subreddit winterporn
...
Selected http://i.imgur.com/3UWbcYG.jpg (EarthPorn) with score 5.21729920261845e-05 out of 5971 images
The probability of selecting this image was 0.009851421028579594
Applying wallpaper

Scoring algorithm

Images are scored in three steps as follows:

1. Three input scores are calculated for each image in the cache

   
   pixel_score = [totalimagepixels]/[totalscreenpixels]
   
   ratio_score = [imagex − yratio]/[screenx − yratio]or[screenx − yratio]/[imagex − yratio](whicheverislessthan1)
   
   $$views_score = [# views of this image] / [highest # views of all images]$$
   

2. Each input score is run through a sigmoid function, in this case, the logistic function. This helps to provide stronger differentiation between a good and a bad image than could be afforded with a linear method. The logistic function is defined by its midpoint and the slope at the midpoint, k.

   

   Logistic function with midpoint=0, k=1

   In redrum the logistic function has been normalized such that f(0) = 1.

   
   ratio_logistic_score = (1 + pow(math.e,  − ratio_k * (1 − ratio_off)))/(1 + pow(math.e,  − ratio_k * (ratio_score − ratio_cutoff)))views_logistic_score = (1 + pow(math.e,  − ratio_k * (1 − ratio_off)))/(1 + pow(math.e,  − views_k * (views_score − views_cutoff)))pixel_logistic_score = (1 + pow(math.e,  − ratio_k * (1 − ratio_off)))/(1 + pow(math.e,  − pixel_k * (pixel_score − pixel_cutoff)))
   

   pixel_score, ratio_score, and views_score each have their own midpoint and k, which can be set in ~/.config/redrum.ini

3. The three logistic scores are then multiplied to calculate an image's final_score. This score is used to do a random weighted select of all images in the cache. An image with a final_score that is twice the score of another image is twice as likely to be selected during the random selection.

Tuning the scoring algorithm

You can adjust the scoring algorithm if you aren't satisfied with the images being selected. redrum_tune allows you to quickly adjust the midpoint and k of the logarithmic function and view its effects on the final_score. You need to run pip install redrum[tune] to install the extra dependencies (matplotlib, numpy).