How ani3h/semcom works
semcom is a semantic-communication research package: its current runtime supports modality-specific dataset loading, transceiver training through simulated channels, task-metric evaluation across SNR values, and experiment reporting. The main workflow is the Week 4 training entry point, which instantiates configured task, channel, and datasets, trains and evaluates, then saves results and plots. The README also documents audio, image, and text tasks plus AWGN/Rayleigh comparisons. Some dataset and modality wiring is configured dynamically; the graph shows only relationships evidenced by the README, source index, or excerpts.
Main entry points: week4_train.py Training driver · encoders/base.py Task interface · data/base.py Dataset interface · task.py Audio task
Components 25
Data loading
- Dataset interface
src/semcom/data/base.py - Dataset splitting
src/semcom/data/splits.py - Edinburgh audio
src/semcom/data/audio_edinburgh.py - CIFAR-10 and MNIST
src/semcom/data - Europarl text
src/semcom/data/text_europarl.py
Transceivers and channels
- Task interface
src/semcom/encoders/base.py - Audio task
src/semcom/encoders/audio/task.py - Audio transceiver
src/semcom/encoders/audio/model.py - Image tasks
src/semcom/encoders/image/task.py - Image transceiver
src/semcom/encoders/image/model.py - Text task
src/semcom/encoders/text/task.py - Text transceiver
src/semcom/encoders/text/model.py - Word tokenizer
src/semcom/encoders/text/tokenizer.py - Channel interface
src/semcom/channel/base.py - AWGN simulation
src/semcom/channel/awgn.py - Rayleigh simulation
src/semcom/channel/rayleigh.py
Training and evaluation
- Training driver
experiments/week4_train.py - Experiment config
configs/week4.yaml - Audio metrics
src/semcom/encoders/audio/metrics.py - Metrics and artifacts
experiments/week4_train.py
Experiment tracking
- Run naming
src/semcom/tracking/run_naming.py - Random seeding
src/semcom/tracking/seed.py - W&B logging
src/semcom/tracking/wandb_logger.py - SNR plots
src/semcom/tracking/plots.py
Other components
- Experiment user
Connections 29
- Experiment user → Training driver: runs
README.md - Training driver → Experiment config: loads config
experiments/week4_train.py - Training driver → Task interface: runs task methods
experiments/week4_train.py - Training driver → Channel interface: instantiates channel
experiments/week4_train.py - Training driver → Dataset interface: instantiates datasets
experiments/week4_train.py - Edinburgh audio → Dataset interface: implements
src/semcom/data/audio_edinburgh.py - Edinburgh audio → Dataset splitting: gets split indices
src/semcom/data/audio_edinburgh.py - CIFAR-10 and MNIST → Dataset interface: uses interface
src/semcom/data/image_cifar10.py - CIFAR-10 and MNIST → Dataset splitting: uses split indices
src/semcom/data/image_cifar10.py - Europarl text → Dataset interface: uses interface
src/semcom/data/text_europarl.py - Europarl text → Dataset splitting: uses split indices
src/semcom/data/text_europarl.py - Audio task → Task interface: implements
src/semcom/encoders/audio/task.py - Audio task → Audio transceiver: builds model
src/semcom/encoders/audio/task.py - Audio task → Audio metrics: computes metrics
src/semcom/encoders/audio/task.py - Image tasks → Task interface: implements
src/semcom/encoders/image/task.py - Image tasks → Image transceiver: builds model
src/semcom/encoders/image/task.py - Text task → Task interface: implements
src/semcom/encoders/text/task.py - Text task → Text transceiver: builds model
src/semcom/encoders/text/task.py - Text task → Word tokenizer: batches text
src/semcom/encoders/text/task.py - Audio transceiver → Task interface: implements transceiver
src/semcom/encoders/audio/model.py - Text transceiver → Task interface: implements transceiver
src/semcom/encoders/text/model.py - Text transceiver → Word tokenizer: uses token IDs
src/semcom/encoders/text/model.py - AWGN simulation → Channel interface: implements
src/semcom/channel/awgn.py - Rayleigh simulation → Channel interface: implements
src/semcom/channel/rayleigh.py - Training driver → Random seeding: sets seed
experiments/week4_train.py - Training driver → Run naming: names run
experiments/week4_train.py - Training driver → W&B logging: starts and logs run
experiments/week4_train.py - Training driver → SNR plots: plots SNR curves
experiments/week4_train.py - Training driver → Metrics and artifacts: writes evaluation results
experiments/week4_train.py