"""Chart embedding with a Columnist pretraining checkpoint (needs columnist.py next to this file). from embed_infer import load_encoder, embed_chart enc = load_encoder("c_big.pt") vec = embed_chart(enc, tok, gap, sv) # (d,) numpy vector; tok/gap/sv are the arrays of one 4K chart (see README.md) """ import numpy as np import torch import columnist as C def load_encoder(path, device=None): dev = torch.device(device or ("cuda" if torch.cuda.is_available() else "cpu")) ck = torch.load(path, map_location="cpu", weights_only=False) a = ck["args"] m = C.Columnist(a["d"], a["layers"], look=bool(a.get("look", False)), v3=bool(a.get("v3_inputs", False))) m.load_state_dict(ck["model"], strict=False) # the pretraining checkpoints also hold optional heads return m.to(dev).eval() @torch.no_grad() def embed_chart(enc, tok, gap, sv, L=512): """Mean over the valid rows of the encoder output of each non-overlapping L-row window, averaged over windows weighted by their length.""" dev = next(enc.parameters()).device tok = np.asarray(tok).astype(np.int64); gap = np.clip(np.nan_to_num(np.asarray(gap, dtype=np.float32)), 0, 60000); sv = np.asarray(sv, dtype=np.float32) n = len(tok); starts = list(range(0, max(1, n), L)); W = len(starts) T = np.zeros((W, L), np.int64); G = np.zeros((W, L), np.float32); S = np.ones((W, L), np.float32); V = np.zeros((W, L), bool) for k, s in enumerate(starts): m = min(L, n - s); T[k, :m], G[k, :m], S[k, :m], V[k, :m] = tok[s:s + m], gap[s:s + m], sv[s:s + m], True T, G, S, V = (torch.from_numpy(x).to(dev) for x in (T, G, S, V)) e = C.Columnist.pool(enc(T, G, S, V).float(), V) # (W, d) w = V.sum(1).float() return ((e * w[:, None]).sum(0) / w.sum()).cpu().numpy()