| # ailab-bio/TACK-Context-Encoding | |
| PCA-reduced context embedding tables for the `tackai` fusion surrogates (`tackai.fusion`): | |
| one lookup table plus a PCA side-file per block (assay (8-d), cell (49-d), e3 (7-d), poi (51-d)), and `manifest.json` recording | |
| each file's sha256, byte size and dimensions. Published from a `protein_space="combined"` encoder; `from_pretrained` reproduces that by default. | |
| These tables are **inputs** to every fusion model, never the output of one — a fitted | |
| ensemble is only meaningful against the exact tables it was built with, so this repo is | |
| versioned by commit rather than by file content. | |
| ## Loading | |
| ```python | |
| from tackai.fusion.data import FusionData | |
| data = FusionData.from_pretrained("ailab-bio/TACK-Context-Encoding") | |
| context = data.transform_context({"poi_seq": ..., "e3_seq": ..., "cell_id": ..., "assay": ...}) | |
| ``` | |
| `from_pretrained` installs these files into `TACKAI_CACHE` and verifies them against this | |
| repo's `manifest.json` before use. | |
| Generated by `tackai.fusion.context.stage_context_tables` (tackai version | |
| 1.0.0) — do not hand-edit; re-stage and re-upload instead. | |