Learning Distance Estimators from Pivoted Embeddings of Metric Objects

Code to reproduce the results presented in 'Learning Distance Estimators from Pivoted Embeddings of Metric Objects'.

How to reproduce

1. Get and prepare the data (instructions here)

2. pip install -r requirements.txt

3. python train_all.py yfcc100m-hybridfc6

The train_all.py script adopts Ray Tune to handle and parallelize runs.
You should modify the script and adjust the parameters in ray.init() and tune.run() to your resources and desired parallelism.