A benchmarking framework for neural decoding from podcast listening data.
Comparing brain → word decoding performance to previously published results.
📚 Full documentation available at: https://hassonlab.github.io/podcast-benchmark/
# Setup environment and download data
./setup.sh
# Train word embedding decoding task
make train-all TASKS=word_embedding_decoding_task
# Train all tasks over all models
make train-all- Flexible model architecture: Register custom models with simple decorators
- Multiple tasks: Word embeddings, classification, or custom prediction targets
- Configurable training: YAML-based configs with cross-validation and early stopping
- Multiple metrics: ROC-AUC, perplexity, top-k accuracy, and custom metrics
- Time lag analysis: Automatically find optimal temporal offsets
- Quickstart Guide - Get up and running
- Onboarding a Model - Add your own models
- Adding a Task - Create custom tasks
- Configuration - Understanding configs
- Registry API - Function signatures