How y4h1ko/signal-inference-transformers-normalising-flows works
This research code generates synthetic sinusoidal observations and trains encoder-only Transformers to infer their frequencies. The principal workflow is configured data generation, tensor conversion and splitting, model training, test evaluation, then plots and posterior diagnostics. It covers single-frequency regression and conditional-flow inference, plus linear and nonlinear two-frequency mixtures with permutation ambiguity. The graph also preserves positional encoding, experiment-analysis utilities, and plotting as distinct internal capabilities. Training and several model details are represented from README and call sites; their source excerpts were not provided, so those links are limited to visible calls or documentation.
Main entry points: main.py Experiment runner · models.py Transformer encoders · imports_and_libraries.py Shared runtime imports · dataset_creation.py Signal datasets
Components 15
Inference workflow
- Experiment runner
main.py - Signal datasets
code_files/dataset_creation.py - Tensor conversion and splits
code_files/dataset_creation.py - Training loops
code_files/train_and_test.py - Test evaluation
code_files/train_and_test.py
Modeling
- Transformer encoders
code_files/models.py - Positional encoding
code_files/positional_encodings.py - Deterministic regression
code_files/models.py - Single-frequency flow
code_files/models.py - Two-frequency flow
code_files/models.py
Evaluation and analysis
- Plots and diagnostics
code_files/visualizations.py - Experiment sweeps
archive/analysis.py
Configuration
- Experiment settings
code_files/configuration.py - Shared runtime imports
code_files/imports_and_libraries.py
Other components
- Researcher
Connections 15
- Researcher → Experiment runner: runs experiments
README.md - Experiment runner → Signal datasets: generates signals
main.py - Experiment runner → Tensor conversion and splits: converts and splits
main.py - Experiment runner → Training loops: trains models
main.py - Experiment runner → Test evaluation: evaluates models
main.py - Experiment runner → Plots and diagnostics: plots results
main.py - Experiment runner → Deterministic regression: selects regression
main.py - Experiment runner → Single-frequency flow: selects flow model
main.py - Experiment runner → Two-frequency flow: selects pair model
main.py - Transformer encoders → Positional encoding: adds positions
code_files/models.py - Deterministic regression → Transformer encoders: uses encoder
README.md - Single-frequency flow → Transformer encoders: conditions on encoder
README.md - Two-frequency flow → Transformer encoders: conditions on encoder
README.md - Experiment runner → Experiment sweeps: analyzes settings
- Shared runtime imports → Experiment settings: loads settings
code_files/imports_and_libraries.py