log.to_pandas() gives only NaNs

[RiccPicc]

Hi everyone! Can you help me? I am new in using DeepSurv. I am following the steps given in the examples, but I encounter this issue in the dataset on which I wish to use this package:

Before running the net, I ensured that all the variables I am using were float. I converted categories into numeric values using .cat.codes and then converted into float. I did so since otherwise model_.lr_finder(x_train, y_train, batch_size, tolerance=10) and model_.fit(x_train, y_train, batch_size, epochs, callbacks, verbose, val_data=val, val_batch_size=batch_size) wouldn't have worked due to different "data" type.

This are the setting I am using for the tt.practical.MLPVanilla(in_features, num_nodes, out_features, batch_norm, dropout, output_bias=output_bias):

n_nodes = 256
in_features = x_train.shape[1] # number of variables
num_nodes = [n_nodes, n_nodes, n_nodes, n_nodes]
out_features = 1
batch_norm = True 
dropout = 0.4 
output_bias = False

This is the model: model_ = CoxPH(net_ds, tt.optim.Adam)

This are fit settings for log = model_.fit(x_train, y_train, batch_size, epochs, callbacks, verbose,val_data=val, val_batch_size=batch_size):

batch_size = 128
best_lr = lrfinder.get_best_lr()
model_.optimizer.set_lr(best_lr)
epochs = 512
callbacks = [tt.callbacks.EarlyStopping()] # Stop training when a monitored metric has stopped improving.
verbose = True

Thank you in advance! Cheers