DEP/create-dataset.py at main · SnowCharmQ/DEP · 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
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
import os
import torch
import pandas as pd
from tqdm import tqdm
from collections import defaultdict
from transformers import AutoTokenizer, set_seed
from datasets import load_dataset, concatenate_datasets, Dataset

from data.personal_dataset import convert_to_dataset, PersonalDataset

set_seed(30)

if not os.path.exists("output"):
    os.makedirs("output")

categories = ["Books", "Movies_and_TV", "CDs_and_Vinyl"]
dataset_len = [317, 1925, 1754]

train_datasets = []
meta_datasets = []
for i, category in enumerate(categories):
    train_dataset = load_dataset(
        "SnowCharmQ/DPL-main",
        category,
        split="train"
    ).map(lambda _: {"category": category})
    df = pd.DataFrame(train_dataset)
    df['profile_length'] = df['profile'].apply(len)
    dataset = (
        df.sort_values('profile_length', ascending=False)
        .groupby('user_id')
        .head(1)
        .reset_index(drop=True)
    )
    train_dataset = Dataset.from_pandas(dataset)
    train_datasets.append(train_dataset)
    meta_dataset = load_dataset(
        "SnowCharmQ/DPL-meta",
        category,
        split="full"
    )
    meta_datasets.append(meta_dataset)

train_dataset = concatenate_datasets(train_datasets)
train_dataset = train_dataset.shuffle(seed=42)
meta_dataset = concatenate_datasets(meta_datasets)
meta_dataset = dict(zip(meta_dataset["asin"],
                        zip(meta_dataset["title"],
                            meta_dataset["description"])))

val_datasets = []
for i, category in enumerate(categories):
    val_main_dataset = load_dataset(
        "SnowCharmQ/DPL-main",
        category,
        split="val"
    ).map(lambda _: {"category": category})
    val_datasets.append(val_main_dataset)
val_dataset = concatenate_datasets(val_datasets)
val_dataset = val_dataset.shuffle(seed=42)
val_dataset = val_dataset.select(range(512))

test_datasets = []
for category in categories:
    test_dataset = load_dataset(
        "SnowCharmQ/DPL-main",
        category,
        split="test"
    ).map(lambda _: {"category": category})
    test_datasets.append(test_dataset)
test_dataset = concatenate_datasets(test_datasets)

llm_model_name = "Qwen/Qwen2.5-7B-Instruct"
llm_tokenizer = AutoTokenizer.from_pretrained(llm_model_name)
llm_tokenizer.padding_side = "left"
new_tokens = [f"[HIS_TOKEN_{i}]" for i in range(8)] + \
    [f"[DIFF_TOKEN_{i}]" for i in range(8)] + \
    ["<his_token_start>", "<his_token_end>",
        "<diff_token_start>", "<diff_token_end>"]
llm_tokenizer.add_special_tokens({"additional_special_tokens": new_tokens})
llm_tokenizer.save_pretrained("output/DEP-tokenizer")

user_his_emb_map = {}
user_prof_mean_emb_map = {}
asin_reviewers_map = defaultdict(set)
for sample in tqdm(test_dataset, desc="Pre-Processing the dataset"):
    user_id = sample["user_id"]
    category = sample["category"]
    profile = sample["profile"]
    profile = profile[:-2]
    his_emb = torch.load(f"embeddings/{category}/{user_id}.emb", weights_only=True)
    his_emb = his_emb[:-2]
    prof_mean_emb = torch.mean(his_emb, dim=0)
    user_his_emb_map[f"{user_id}_{category}"] = his_emb
    user_prof_mean_emb_map[f"{user_id}_{category}"] = prof_mean_emb
    for i, p in enumerate(profile):
        asin_reviewers_map[p["asin"]].add((user_id, i))

personal_dataset = PersonalDataset(
    train_dataset,
    meta_dataset,
    user_his_emb_map,
    user_prof_mean_emb_map,
    asin_reviewers_map,
    llm_tokenizer=llm_tokenizer,
    new_tokens=new_tokens,
    training=True
)
hf_dataset = convert_to_dataset(personal_dataset)
hf_dataset.save_to_disk("data/dataset_train")

personal_dataset = PersonalDataset(
    val_dataset,
    meta_dataset,
    user_his_emb_map,
    user_prof_mean_emb_map,
    asin_reviewers_map,
    llm_tokenizer=llm_tokenizer,
    new_tokens=new_tokens,
    training=False
)
hf_dataset = convert_to_dataset(personal_dataset)
hf_dataset.save_to_disk("data/dataset_val")

for i, category in enumerate(categories):
    test_train_dataset = load_dataset(
        "SnowCharmQ/DPL-main",
        category,
        split="test"
    ).map(lambda _: {"category": category})
    test_dataset = PersonalDataset(
        test_train_dataset,
        meta_dataset,
        user_his_emb_map,
        user_prof_mean_emb_map,
        asin_reviewers_map,
        llm_tokenizer=llm_tokenizer,
        new_tokens=new_tokens,
        training=False
    )
    hf_dataset = convert_to_dataset(test_dataset)
    hf_dataset.save_to_disk(f"data/dataset_test_{category}")