Hyper parameters of blogcatalog for SDNE

[agoodweathercc]

Hi,

I am having trouble reproducing the result of blogcatalog using SDNE. It's similar to the issue mentioned here.
#35

would it be possible for you to share the hyperparameter for blogcatalog dataset?

Thanks a lot!

[palash1992]

Here is the hyperparameter range which was tested for BlogCatalog:
blogcat_hypRange_1.txt

[palash1992]

[agoodweathercc]

Hello,

Thanks for providing the hyper-parameter. I found that for SDNE

{'alpha': [1e-05],
'beta': [2, 5, 10, 20],
'modelfile': [['gem/intermediate/enc_model.json',
'gem/intermediate/dec_model.json']],
'n_batch': [500],
'n_iter': [100],
'n_units': [[], [128], [512, 128], [1024, 512, 128]],
'nu1': [0.0001],
'nu2': [0.0001],
'rho': [0.3],
'weightfile': [['gem/intermediate/enc_weights.hdf5',
'gem/intermediate/dec_weights.hdf5']],
'xeta': [0.01]}

Since there are 4 values for beta and 4 for n_units, there are still 16 combinations. I was wondering am I suppose to train 16 models separately and then evaluate them one by one? Or it can be handled internally? (Only best model will be saved)