Vali Loss: nan Test Loss: nan

[ezelanza]

When I run exp.train(setting), with my own data. I've got

Epoch: 1, Steps: 35 | Train Loss: 1.0262758 Vali Loss: nan Test Loss: nan

I'm trying to debug it but I can't see what is the problem?, the data is the same, so why am I getting a train_loss value while valid_loss = nan, when is the "same data"

I've tried with multiple normalizations, sometimes I can get values but the predictions are not good at the end. Would you help me to understand how the data should be prepared or where is the problem? I'm attaching my dataset, it has 60 features (it's already normalized)

Any comments ? Train_norm_full.csv

[Med-Rokaimi]

I had the same problem. The problem was because I deleted values of some rows from the bottom of the csv files but it still read the as nan values. so make sure to delete the raws instead of only values

Try to increase the dataset size and reducing the batch size as well.

[MountVoom]

I took a look at your data and the value is very small and it's almost zero. This may cause one or more of the columns to have a variance of 0. Thus nan appears on normalization. You can check if nan exists in the data you are using, and if the variance is 0 when normalized.

[AnTi-anti]

数据中包含缺失值nan