Hugging Face's logo

ChatterjeeLab
/
pCoMole

Model card Files
xet
 Community
pCoMole
File size: 4,255 Bytes
12fea4a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
import os
import random
import pandas as pd
from datasets import Dataset, DatasetDict
from transformers import EsmTokenizer

import selfies as sf
from rdkit import Chem
from rdkit import RDLogger
RDLogger.DisableLog('rdApp.*')

import sys
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from smiles_tokenizer.my_tokenizers import SMILES_SPE_Tokenizer
from smiles_tokenizer.selfies_tokenizers import SelfiesTokenizer

# ---------------------------------------------------------------------
# CONFIG
# ---------------------------------------------------------------------
PROJECT_ROOT = os.environ.get('PROJECT_ROOT', 'path/to/project')
CSV_PATH = os.path.join(PROJECT_ROOT, "data/cas9/cas9.csv")
OUTPUT_DIR = os.path.join(PROJECT_ROOT, "data/cas9")
MAX_TOKENS_PER_BATCH = 8192
TRAIN_RATIO, VAL_RATIO, TEST_RATIO = 0.8, 0.1, 0.1


# ------------------------------------------------------------
# 1) Load CSV and split 8:1:1
# ------------------------------------------------------------
df = pd.read_csv(CSV_PATH)
sequences = df["Sequence"].tolist()

# ---- Build vocab ----
tokenizer = EsmTokenizer.from_pretrained("facebook/esm2_t33_650M_UR50D")

random.shuffle(sequences)
n = len(sequences)
n_train = int(n * TRAIN_RATIO)
n_val = int(n * VAL_RATIO)

train_df = sequences[:n_train]
val_df   = sequences[n_train:n_train + n_val]
test_df  = sequences[n_train + n_val:]


# ------------------------------------------------------------
# 2) Tokenizer
# ------------------------------------------------------------

def tokenize_selfies(sequences):
    enc = {"input_ids": [], "attention_mask": []}
    for s in sequences:
        res = tokenizer(s)
        enc["input_ids"].append(res["input_ids"])
        enc["attention_mask"].append(res["attention_mask"])
    # enc["input_ids"] is a list of list[int]
    return enc


# ------------------------------------------------------------
# helper: build batches (one row = one batch)
# ------------------------------------------------------------
def build_batched_dataset(sequences, max_tokens=9182):
    # tokenize all first
    toks = tokenize_selfies(sequences)
    input_ids_list = toks["input_ids"]
    attn_mask_list = toks["attention_mask"]

    # add lengths
    items = []
    for ids, mask in zip(input_ids_list, attn_mask_list):
        items.append({
            "input_ids": ids,
            "attention_mask": mask,
            "length": len(ids),
        })

    # sort by length
    items.sort(key=lambda x: x["length"])

    batched_input_ids = []
    batched_attention_masks = []
    batched_lengths = []
    batched_batch_sizes = []

    i = 0
    n = len(items)
    while i < n:
        L = items[i]["length"]
        # how many of length L can we pack?
        max_bs = max_tokens // L
        if max_bs < 1:
            max_bs = 1

        # collect up to max_bs items with same length
        cur_ids = []
        cur_masks = []
        taken = 0
        j = i
        while j < n and items[j]["length"] == L and taken < max_bs:
            cur_ids.append(items[j]["input_ids"])         # length L
            cur_masks.append(items[j]["attention_mask"])  # length L
            j += 1
            taken += 1

        # now this is one batch: shape (B, L)
        batched_input_ids.append(cur_ids)
        batched_attention_masks.append(cur_masks)
        batched_lengths.append(L)
        batched_batch_sizes.append(taken)

        i = j

    ds = Dataset.from_dict({
        "input_ids": batched_input_ids,           # list of (B, L)
        "attention_mask": batched_attention_masks,
        "seq_length": batched_lengths,
        "batch_size": batched_batch_sizes,
    })
    return ds


# ------------------------------------------------------------
# 3) build batched datasets for each split
# ------------------------------------------------------------
train_ds = build_batched_dataset(train_df, MAX_TOKENS_PER_BATCH)
val_ds   = build_batched_dataset(val_df,   MAX_TOKENS_PER_BATCH)
test_ds  = build_batched_dataset(test_df,  MAX_TOKENS_PER_BATCH)

dsdict = DatasetDict({
    "train": train_ds,
    "validation": val_ds,
    "test": test_ds,
})

os.makedirs(OUTPUT_DIR, exist_ok=True)
dsdict.save_to_disk(OUTPUT_DIR)
print(f"saved to {OUTPUT_DIR}")