File size: 2,807 Bytes
db32e07
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
"""
Optional SMI-TED encoder for molecule embeddings (E_smi).
Uses De-SpecBridge's load_smited when DESPECBRIDGE_PATH is set or --despecbridge-path is given.
"""
from __future__ import annotations

import os
import sys
from pathlib import Path
from typing import List, Optional

import numpy as np


def _l2_normalize(x: np.ndarray, eps: float = 1e-8) -> np.ndarray:
    norm = np.linalg.norm(x, axis=-1, keepdims=True)
    return (x / (norm + eps)).astype(np.float32)


def load_smited_encoder(
    despecbridge_path: Optional[str] = None,
    model_name: str = "ibm-research/materials.smi-ted",
    device: Optional[str] = None,
):
    """
    Load SMI-TED encoder. If despecbridge_path is set, prepend to sys.path and import
    load_smited from despecbridge.models.smited_decoder.
    Returns a callable encode(smiles_list) -> np.ndarray (normalized), or None if unavailable.
    """
    path = despecbridge_path or os.environ.get("DESPECBRIDGE_PATH")
    if path:
        path = str(Path(path).resolve())
        if path not in sys.path:
            sys.path.insert(0, path)
    try:
        from despecbridge.models.smited_decoder import load_smited
    except ImportError:
        return None

    import torch
    dev = device or ("cuda" if torch.cuda.is_available() else "cpu")
    wrapper = load_smited(model_name=model_name, device=torch.device(dev))
    wrapper.eval()

    def encode(smiles: List[str], batch_size: int = 32, desc: Optional[str] = None) -> np.ndarray:
        from tqdm import tqdm
        out = []
        n_batches = (len(smiles) + batch_size - 1) // batch_size
        it = range(0, len(smiles), batch_size)
        if desc:
            it = tqdm(it, total=n_batches, desc=desc, unit="batch")
        for i in it:
            batch = smiles[i : i + batch_size]
            if not batch:
                continue
            # Use encode_mean_pool for [B, D] tensor; .encode() returns a dict.
            if hasattr(wrapper, "encode_mean_pool"):
                with torch.no_grad():
                    h = wrapper.encode_mean_pool(batch)
            else:
                with torch.no_grad():
                    raw = wrapper.encode(batch)
                h = raw["last_hidden_state"] if isinstance(raw, dict) else raw
                if h.dim() == 3:
                    h = h.mean(dim=1)
            if isinstance(h, torch.Tensor):
                h = h.cpu().numpy()
            h = np.asarray(h, dtype=np.float32)
            if h.ndim == 0:
                h = np.expand_dims(np.expand_dims(h, 0), 0)
            elif h.ndim == 1:
                h = h.reshape(1, -1)
            out.append(h)
        if not out:
            return np.zeros((0, 0), dtype=np.float32)
        arr = np.concatenate(out, axis=0).astype(np.float32)
        return arr

    return encode