DeepTimeGeo

The official PyTorch implementation of DeepTimeGeo: Trajectory Reconstruction from Sparse Data with Transformer.

⚙️ Installation

• Data
  • Download the Los Angeles dataset from here and move it to DeepTimeGeo/data/los_angeles/

🚀 Running

• The model can be accssed at python train_model.py

• Please find the output at trained_models/. asset/parse.ipynb has the starter code to convert the rank-based representation back to geospatial data. Within trained_models/:

  • data/ contains the processed rank-based representation of the trajectories. training_set.pkl contains trajectories used in training. validation_set.pkl contains users whose trajectories have been procssed but not used in training.
  • simulation/ contains the completed trajectories. User asset/parse.ipynb to convert the rank-based representation back to original.
  • viz/ contains basic visualizations of human moibility patterns in the sparse and the completed trajectories.