SecretoGen/compute_perplexity_spgen.py at main · fteufel/SecretoGen · GitHub

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
import torch
import sys
import pandas as pd
sys.path.append('..')
from src.baselines.datasets import SPGenDataset
from tqdm.auto import tqdm
import numpy as np
import argparse
import os

device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')

from src.baselines.transformer import Models
from src.baselines.progen.modeling_progen import ProGenForCausalLM
from tokenizers import Tokenizer

from fairscale.nn.data_parallel import FullyShardedDataParallel as FSDP
from fairscale.nn.wrap import enable_wrap, wrap
url = "tcp://localhost:12355"
torch.distributed.init_process_group(backend="nccl", init_method = url, world_size=1, rank=0)

spgen_checkpoints = {
    75: '../../SPGen/remote_generation/signal_peptide/outputs/SIM75_550_12500_64_6_5_0.1_64_100_0.0001_-0.03_99_weightsonly.pt',
    90: '../../SPGen/remote_generation/signal_peptide/outputs/SIM90_550_12500_64_6_5_0.1_64_100_0.0001_-0.03_99_weightsonly.pt',
    95: '../../SPGen/remote_generation/signal_peptide/outputs/SIM95_550_12500_64_6_5_0.1_64_100_0.0001_-0.03_99_weightsonly.pt',
    99: '../../SPGen/remote_generation/signal_peptide/outputs/SIM99_550_12500_64_6_5_0.1_64_100_0.0001_-0.03_99_weightsonly.pt',
}

def load_spgen_model(checkpoint: int = 99):
    state_dict = torch.load(spgen_checkpoints[checkpoint])
    model = Models.Transformer(
        27,
        27,
        107,
        proj_share_weight=True,
        embs_share_weight=True,
        d_k=64,
        d_v=64,
        d_model=550,
        d_word_vec=550,
        d_inner_hid=1100,
        n_layers=6,
        n_head=5,
        dropout=0.1)

    model.load_state_dict(state_dict)
    model.eval()

    return model

# Get the alphabets and add in a padding character (' '), a stop character ('.'), and a start character ('$').
# with open('../data/ctable_copies/ctable_token_master.pkl', 'rb') as f:
#     ctable = pickle.load(f)
def get_perplexity_batch(transformer, src_seq, src_positions, tgt_seq, tgt_positions):
    '''Adapted from Translator()._epoch().'''
    ppls = []

    loss_fn = torch.nn.CrossEntropyLoss()

    pred = transformer((src_seq, src_positions), (tgt_seq, tgt_positions))

    # process each sample in batch
    for idx in range(len(src_seq)):
        loss = loss_fn(pred[idx].view(-1, 27), tgt_seq[idx,1:].view(-1))
        ppls.append(torch.exp(loss).item())

    return ppls


def predict_spgen(model, loader):

    all_aa_logits = []
    all_aa_targets = []

    with torch.no_grad():

        ppl = []
        for idx, batch in tqdm(enumerate(loader), total=len(loader)):


            proteins, prot_positions, sps, sp_positions = batch
            proteins, prot_positions, sps, sp_positions = proteins.to(device), prot_positions.to(device), sps.to(device), sp_positions.to(device)

            aa_logits = model((proteins,prot_positions), (sps, sp_positions)) #sp_len -1, last pos trimmed

            ppls = get_perplexity_batch(model, proteins, prot_positions, sps, sp_positions)

            ppl.extend(ppls)

    return np.array(ppl)


def main():

    parser = argparse.ArgumentParser()
    # parser.add_argument('checkpoint')
    parser.add_argument('out_dir')
    parser.add_argument('--data', type=str, default = '../data/fitness_data/grasso_prepared.csv')
    parser.add_argument('--checkpoint', type = int, default=99)
    args = parser.parse_args()


    model = load_spgen_model(args.checkpoint)
    model.to(device)
    os.makedirs(args.out_dir, exist_ok=True)

    test_set = SPGenDataset(args.data)


    loader = torch.utils.data.DataLoader(test_set, collate_fn = test_set.collate_fn, batch_size=500)
    perplexities = predict_spgen(model, loader)

    df = test_set.df
    df['perplexity'] = perplexities
    df.to_csv(os.path.join(args.out_dir, args.data.split('/')[-1]))


if __name__ == '__main__':
    main()