random_seed does not seem to work

[adradr]

Hey there! 👋🏻

First, I'd like to thank this library as it works like a charm on CUDA. Second I have stumbled upon an issue that I am not totally understanding regarding fixing a random_seed. The sklean version uses a parameter called random_state so I suppose this would be for the same purpose, however when setting like the following I receive a different result when executing the same code twice on the exactly same dataset.

When doing the exactly same with the sklearn.manifold.TSNE imported then I am getting the same result using a fixed randomness when executing twice. Also getting different results using sklearn vs tsne-cuda

Why is that, what am I missing?

Thanks, adr

# Using t-SNE for dimensionality reduction
tsne = TSNE(n_components=2, random_seed=420, perplexity=30, n_iter=1000, verbose=1)
tsne_results = tsne.fit_transform(X)

[JackieQiang]

same question

[callzhang]

any updates?

[ethahtz]

same question