Problem Description

Hi, I got some questions with this code. I converted the part0_train and part1_test parts of the MS1M data set into npy files as you described in the readme. The feature files were directly converted into npy files after l2 regularization and label file reading. And I followed the steps suggested by the readme. In the end, I only calculated the pairwise indicator (sim_thres was set to increase from 0.5 to 0.8 in steps of 0.05), but the results kept prompting that ave_pre was 0, ave_rec and fscore were nan. , I would like to ask how to solve this situation?

By the way, when running the faiss_search.sh file, since my GPU is not compatible with faiss-gpu, I used faiss-cpu to solve this problem.

Environment

GPU A100
CUDA 11.0
Python3.6 or Python 3.9
torch 1.7.1+cu110 or torch 1.10.1

使用http://idstcv.oss-cn-zhangjiakou.aliyuncs.com/Ada-NETS/MSMT17/msmt17_feature_label.zip给出的数据集跑不出论文中的结果
top改为40的时候bcubed: ave_pre: 0.8325, ave_rec: 0.6552, fscore: 0.7333, cluster_num: 15176
top改为80的时候bcubed: ave_pre: 0.7136, ave_rec: 0.6641, fscore: 0.6879, cluster_num: 14958, singleton_num: 11450
并且cluster.py文件中的cos_sim_thres和sim_thres都没有给出定义
请问应该如何修改

在论文中的等式2中，定义了新的相似度等于jaccard相似度和cosine相似度的加权和，使用到参数η。但是我在论文中并没有找到实验时这个参数的定义，在代码中发现一个参数lamb=0.3，但是加权的是距离。

所以代码中的lamb=0.3相当于论文中的η=0.7？

作者，您好，关于struct space，如果struct_space能改变特征集的整体结构，那就可以直接很好的聚类或者识别了，所以不太明白后续的自适应的意义有多大？

前辈您好：

    首先恭喜您的团队论文高中

   关于您提供的数据集，在上图数字“2”标注处，下载下来的数据，训练集数据是30248，测试集数据是93820，而上图数字“1”标注处，进入“1”的链接后下载的数据（Part1 (584K): GoogleDrive or BaiduYun (passwd: geq5)），训练集数据是576494，测试集数据是584013。
   在上图数字“2”标注处下载的数据上，也就是训练集数据是30248，测试集数据是93820，最终效果达不到您论文所述，和您论文结果相差很大。想请教您一下，您在训练模型和测试模型的时候，训练集和测试集数量分别设置的是多少？

Dear author,

when I run>>bash script/train_AND.sh

I get: FileNotFoundError: [Errno 2] No such file or directory: 'outpath/ckpt_40000.pth'

all the other necessary folders are created, but the 'AND/outpath' folder is actually empty. I downloaded the pretrained checkpoints but the mentioned one in the error is not there.
Maybe I miss downloading smtn else?

hi, I got the problem shown in the below when running the script"sh script/train_GCN.sh "

Thank you for your great work.
I run the train and test scripts on the ms1m dataset these days,but I got poor performance.
Here are my result on ms1m part1_test:
Bcubed fscore:0.6264, parwise fscore 0.4994.
I wonder which step went wrong，could you share the training log?
When I run the train_GCN script ,the last loss is:
loss:1.9265 bclloss_pos:0.007 bclloss_neg:1.919

论文中相似度的计算是Jaccard距离和cosine相似度的加权和。

但是代码实现好像不一样。如果我没搞错的话这部分是在tool/struct_space.py中实现的，如下

sim = 1.0 * len(query_Rstarset & doc_Rstarset) / len(query_Rstarset | doc_Rstarset)
jd = 1 - sim
cd = D[query_nodeid, idx]
nd = (1-lamb) * jd + lamb * cd

其中D是在tool/faiss_search.py中使用faiss计算得到的，但是这里的D是保存的欧式距离？代码如下，这里是使用欧式距离进行索引查找的。而且我打印出来D的值也是由小到大的。

quantizer = faiss.IndexFlatL2(dim)      # 定义量化器/索引为l2距离(欧式距离)，越小越好
cpu_index = faiss.IndexIVFFlat(quantizer, dim, nlist)
cpu_index.nprobe = nprobe

请问是这样的吗，还是我有什么遗漏之处？

also, when I run >>bash script/gene_adj.sh I got this:
Traceback (most recent call last):
File "tool/gene_adj_adanets.py", line 3, in
from knn import fast_knns2spmat, knns2ordered_nbrs, fast_knns2spmat_adaptivek
File "/home/nigar/Ada-NETS/tool/knn.py", line 289, in
class knn_brute_force(knn):
NameError: name 'knn' is not defined

恭喜您的团队论文高中，请问有公开MSMT17数据集提取后特征的计划吗？我们组也在研究相关工作，十分想follow一下您的工作。祝您生活愉快，工作顺利。

I use custom .npy embeddings and labels of my face image dataset. The structure and namings are the same, but when I run >> sh script/faiss_search.sh I get this error : TypeError: in method 'GpuIndexIVFFlat_train', argument 3 of type 'float const *'
. I guess I extracted embeddings differently from the provided by you. Anyway, any help appreciated. I want to cluster a custom dataset

When running 'sh script/train_GCN.sh', meet questions below. My dgl version is 0.6.1. Have you met this before?

File "train.py", line 58, in collate
block1.create_format_()
AttributeError: 'DGLBlock' object has no attribute 'create_format_'

when executed "pip install -r requirements.txt", Error was shown as below:
ERROR: Could not find a version that satisfies the requirement faiss==1.6.3 (from versions: none)
ERROR: No matching distribution found for faiss==1.6.3
So i noted "#faiss==1.6.3"and"#apex==0.1"(cause for the same issue)，the residual requirements parts were installed successfully.
what to deal with the version issue? are there any alternatives version of faiss and apex?
thanks sincerely!