Spectra-Code

This repository provides the word-to-code dictionary, spectrum-encoding utilities, molecular data table, and supplementary benchmark results for The Natural Coding of Language: SPECTRA is All You Need.

Data

• data/data_word_mapping3_clean.json: the fixed 20,005-entry word-to-code cipher used by the original ten-bin, sum-nine workflow.
• data/qm9_cond4.csv: 127,468 QM9-derived molecular records with ten-bin codes.
• examples/mapping_qm9_exact_examples.csv: exact-code examples connecting the fixed dictionary and QM9 table.
• data/second_round/versioned_dictionary_400_1023.jsonl.gz: all 20,005 original ten-bin, sum-nine word-code pairs, each accompanied by a separate 400-bin, sum-1,023 computed-spectrum assignment. The original cipher remains the W2C lookup.
• data/second_round/table_s4.csv through table_s10.csv: the per-design capacity, collision, reconstruction, molecular-readout, perturbation, and resolution results reported in Supplementary Tables S4-S10. See the table guide.

Code

• scripts/w2c.py: exact lookup in the original fixed vocabulary.
• scripts/spectrum2code.py: original ten-bin spectrum-to-code conversion.
• scripts/expanded_encoding.py: equal-width and training-fitted adaptive binning, fixed-total quantization, and occupancy statistics for the expanded-code comparison.
• scripts/dft_log_utils.py: Gaussian frequency-log parsing and IR broadening utilities.
• scripts/validate_release_assets.py and scripts/validate_second_round.py: checks for the released data assets and supplementary results.
• S2M/: spectrum-to-molecule source modules and model configurations.

Quick Checks

Install the packages in requirements.txt, then run:

python scripts/validate_release_assets.py
python scripts/validate_second_round.py
python -m unittest discover -s tests

The data and methods index describes the relationship between the original cipher and the expanded comparison. Repository access and licensing terms are in NOTICE.md.