Skip to content
A quick test of building doppler image mosaics with Media Cloud topic stories
HTML Python JavaScript CSS
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

MediaCloud + Doppler + Top Image Topic Test

Install the requirements first. Then add a .env file and add a line with your api key: MC_API_KEY=1233423532532523 Excerpted from Leon Yin's Disinfo Doppler []

Running the Script

1 - Extract Images from a Timespan

The first step is to make a list of all the images in a timespan:

  1. pass in the topic_id and timespan for which topic images you want OR edit the default constants in

  2. run python to fetch the images and generate a .json file to feed into the doppler

2 - Download Images & Prep for Analysis

Run the script to download and prep the images for analysis: python data/images-123-4321.json

3 - Generate the scatterplot

  1. If you have run the scripts before, delete (and backup) old files: umap_training_data_1000.csv and encoder_0-5_dist_euclidean_sample_1000.pkl and any previous logits and metadata files with the same name (topic id and timespan).
  2. Generate the logits: python -m doppler.logits data/images-123-4321.json
  3. Generate the scatterplot: python -m doppler.mosaics data/images-123-4321.json

4 - Display images in HTML/D3 according to metadata

  1. If you want to pull images according to Inlink count, FB shares, media source, publish date or other metadata, please refer to the treemaps folder which contains html, js and css for our image treemaps.

Installation Tips

For OSX, to get Torch to work you might need to brew install libomp.

You can’t perform that action at this time.