About the time complexity of attach_memory_bank.py

[TinaChen95]

Hi,

I'm trying to reproduce the results on ljspeech dataset, and I noticed that the code in this section (

naturalspeech/attach_memory_bank.py

Line 132 in ca2f814

kmeans = KMeans(n_clusters=1000, random_state=0, n_init="auto").fit(X)

) takes a very long time to execute. It appears that the shape of X is (7066362, 192), which leads to a high computational complexity when calculating the distances.

I was wondering if you could share with me how much time it takes for this code to run on your machine?
Thanks

[Aliasgarsaifee]

Yeah got same issue, for me it had taken around around 10 hours on machine. It runs on CPU that's why it is more slower. And this time is for n_init=1 which is running 1 iteration equivalent to n_init="auto".

[wen0320]

Hi,do you have follow question:

12399 torch.Size([192, 651])
(7066362, 192)
Traceback (most recent call last):
File "attach_memory_bank.py", line 174, in
centers = k_means(zs)
File "attach_memory_bank.py", line 132, in k_means
kmeans = KMeans(n_clusters=1000, random_state=0, n_init="auto").fit(X)
File "/home/WeiWB/anaconda3/envs/tts/lib/python3.7/site-packages/sklearn/cluster/_kmeans.py", line 1146, in fit
self._check_params(X)
File "/home/WeiWB/anaconda3/envs/tts/lib/python3.7/site-packages/sklearn/cluster/_kmeans.py", line 937, in _check_params
if self.n_init <= 0:
TypeError: '<=' not supported between instances of 'str' and 'int'