Hugging Face's logo

JayceAnova
/
Benchmark-Dual

Model card Files
xet
Community
1
Benchmark-Dual
File size: 6,930 Bytes 
936459a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
 
import os
import collections
import json
import logging
import argparse
import numpy as np
import pandas as pd
import torch
from time import time
from torch import optim
from tqdm import tqdm
import torch.utils.data as data
from torch.utils.data import DataLoader
from index.models.rqvae import RQVAE
# from rq_llama import *
# from index.datasets import EmbDataset
import random

class NpyDataset(data.Dataset):
    def __init__(self, data_path):
        self.data_path = data_path
        self.embeddings = np.load(data_path)
        self.dim = self.embeddings.shape[-1]

    def __getitem__(self, index):
        emb = self.embeddings[index]
        tensor_emb = torch.FloatTensor(emb)
        return tensor_emb

    def __len__(self):
        return len(self.embeddings)

def if_collided(all_indices_str):
    tot_item = len(all_indices_str)
    tot_indice = len(set(all_indices_str.tolist()))
    return tot_item == tot_indice

def get_indices_count(all_indices_str):
    indices_count = collections.defaultdict(int)
    for index in all_indices_str:
        indices_count[index] += 1
    return indices_count

def get_collision_item(all_indices_str):
    index2id = {}
    for i, index in enumerate(all_indices_str):
        if index not in index2id:
            index2id[index] = []
        index2id[index].append(i)
    collision_item_groups = []
    for index in index2id:
        if len(index2id[index]) > 1:
            collision_item_groups.append(index2id[index])
    return collision_item_groups

def parse_args():
    parser = argparse.ArgumentParser(description = "Index")
    parser.add_argument("--item_model_path", type = str, default = "", help = "")
    parser.add_argument("--item_data_path", type = str, default = "", help = "")
    parser.add_argument("--user_model_path", type = str, default = "", help = "")
    parser.add_argument("--user_data_path", type = str, default = "", help = "")
    # parser.add_argument("--save_path", type = str, default = "", help = "")
    parser.add_argument("--device", type = str, default = "cuda:0", help = "gpu or cpu")
    return parser.parse_args()

generate_args = parse_args()
print(generate_args)

device = torch.device(generate_args.device)

# generate item index
ckpt = torch.load(os.path.join(generate_args.item_model_path, 'best_collision_model.pth'), map_location = torch.device('cpu'))
args = ckpt['args']
state_dict = ckpt['state_dict']

data = NpyDataset(generate_args.item_data_path)
data_loader = DataLoader(data, num_workers = args.num_workers, batch_size = 64, shuffle = False, pin_memory = True)
# model = RQVAE(
#     in_dim = data.dim,
#     num_emb_list = args.num_emb_list,
#     e_dim = args.e_dim,
#     layers = args.layers,
#     dropout_prob = args.dropout_prob,
#     bn = args.bn,
#     loss_type = args.loss_type,
#     quant_loss_weight = args.quant_loss_weight,
#     kmeans_init = args.kmeans_init,
#     kmeans_iters = args.kmeans_iters,
#     sk_epsilons = args.sk_epsilons,
#     sk_iters = args.sk_iters,
# )
# model.load_state_dict(state_dict)
# model = model.to(device)
# model.eval()
# print(model)

prefix = ["<a_{}>","<b_{}>","<c_{}>","<d_{}>","<e_{}>"]
postfix = "<p_{}>"

index_table = {}
all_indices = []
all_indices_str = []
with torch.no_grad():
    for x in tqdm(data_loader):
        # indices = model.get_indices(x.to(device), False)
        # indices = indices.view(-1, indices.shape[-1]).cpu().numpy()

        indices = np.random.randint(0, 256, size = (64, 4), dtype = int)
        for index in indices:
            code = []
            for i, ind in enumerate(index):
                code.append(prefix[i].format(int(ind)))

            if str(code) in index_table:
                index_table[str(code)] += 1
            else:
                index_table[str(code)] = 0
            code.append(postfix.format(index_table[str(code)]))

            all_indices.append(code)
            all_indices_str.append(str(code))

all_indices = np.array(all_indices)
all_indices_str = np.array(all_indices_str)

print("All indices number: ", len(all_indices))
print("Max number of conflicts: ", max(get_indices_count(all_indices_str).values()))
print('Re-index number:', max(index_table.values()))

all_indices_dict = {}
for item, indices in enumerate(all_indices.tolist()):
    all_indices_dict[item] = list(indices)
reindex_dict = {'reindex': max(index_table.values())}

item_index_path = os.path.join(generate_args.item_model_path, 'indices.random.item.json')
with open(item_index_path, 'w', encoding = 'utf-8') as f:
    json.dump(all_indices_dict, f)

item_reindex_path = os.path.join(generate_args.item_model_path, 'reindex.random.item.json')
with open(item_reindex_path, 'w', encoding = 'utf-8') as f:
    json.dump(reindex_dict, f)

# generate user index
ckpt = torch.load(os.path.join(generate_args.user_model_path, 'best_collision_model.pth'), map_location = torch.device('cpu'))
args = ckpt['args']
state_dict = ckpt['state_dict']

data = NpyDataset(generate_args.user_data_path)
data_loader = DataLoader(data, num_workers = args.num_workers, batch_size = 64, shuffle = False, pin_memory = True)
# model = RQVAE(
#     in_dim = data.dim,
#     num_emb_list = args.num_emb_list,
#     e_dim = args.e_dim,
#     layers = args.layers,
#     dropout_prob = args.dropout_prob,
#     bn = args.bn,
#     loss_type = args.loss_type,
#     quant_loss_weight = args.quant_loss_weight,
#     kmeans_init = args.kmeans_init,
#     kmeans_iters = args.kmeans_iters,
#     sk_epsilons = args.sk_epsilons,
#     sk_iters = args.sk_iters,
# )
# model.load_state_dict(state_dict)
# model = model.to(device)
# model.eval()
# print(model)

prefix = ['<z-{}>','<y-{}>','<x-{}>','<w-{}>','<v-{}>']

all_indices = []
all_indices_str = []
with torch.no_grad():
    for x in tqdm(data_loader):
        # indices = rqvae.get_indices(x.to(device), False)
        # indices = indices.view(-1, indices.shape[-1]).cpu().numpy()
        indices = np.random.randint(0, 256, size = (64, 4), dtype = int)
        for index in indices:
            code = []
            for i, ind in enumerate(index):
                code.append(prefix[i].format(int(ind)))

            all_indices.append(code)
            all_indices_str.append(str(code))

all_indices = np.array(all_indices)
all_indices_str = np.array(all_indices_str)

print("All indices number: ", len(all_indices))
print("Max number of conflicts: ", max(get_indices_count(all_indices_str).values()))

all_indices_dict = {}
for item, indices in enumerate(all_indices.tolist()):
    all_indices_dict[item] = list(indices)

json_path = os.path.join(generate_args.user_model_path, 'indices.random.user.json')
with open(json_path, 'w', encoding = 'utf-8') as f:
    json.dump(all_indices_dict, f)