reproduce tutorial.ipynb

[yuxiaokang-source]

I have noticed the issue https://github.com/ttgump/scDeepCluster/issues/6, so I have created the same environment as you mentioned in readme.md

After running the tutorial you provided in own machine,

the ARI,NMI is very low and did't achieved perfect result. It seems strange that scDeepCluster is unstable for this simulation data, And I noticed that the tutorial.ipynb you have running is not the version you provided in readme.md, the warning information in the tutorial.ipynb seems you are using python3.7, which is not compatible with the python 3.6.3 you mentioned in main page, can you give me some suggestions? Thank you!

[ttgump]

Hi, did you try the pytorch version? Old keras and tensorflow 1 indeed have the problem you mentioned, different version would generate different results. Thanks.

[yuxiaokang-source]

Hi, did you try the pytorch version? Old keras and tensorflow 1 indeed have the problem you mentioned, different version would generate different results. Thanks.

Thank you for you reply, I have try the pytorch version, it gives the expected result. But I found this example of three group is no sense in my view , because the raw dropout counts is well seperated in the tsne plot of ipynb, It can't prove the function of scDeepCluster, can you provide a more powerful example to prove the effect of scDeepCluster

[ttgump]

You can generate datasets with larger dropout rates, set dropout.mid to a larger value such as 0.5 or 1 or 1.5 in the Splatter function. I mentioned in the methods part that we generated datasets with various dropout rates. You can use those settings as an example. Thanks.