Regarding the issue of model usage

[ZChen470]

Hello, I would like to apply the mTAN model to the interpolation task of sine time series with continuous missing values. The data consists of 100 identical sine signals with 450 time points, and there are continuous missing values over several tens of time points. After training, the model predicts well at the known time points, but performs poorly at the missing time points. If you have any suggestions for adjusting hyperparameters, please let me know.

The model parameters are as follows：

 lr=0.0001, std=0.01, latent_dim=1, rec_hidden=32, gen_hidden=50, embed_time=128, k_iwae=5, save=1, enc='mtan_rnn', dec='mtan_rnn', , seed=0, n=100, batch_size=128, quantization=0.016, classif=False, norm=True, kl=True, learn_emb=True, enc_num_heads=1, dec_num_heads=1, num_ref_points=10, dataset='sine', enc_rnn=True, dec_rnn=True, sample_tp=1.0, only_periodic=None, dropout=0.0, predict='false'