Auc evalutation return always nan

Nobody has claimed this yet.

Hello,

I have an issue when i compute roc auc, ti's return always NAN for all the value in the matrix.
`interactions=interactions.tocsr().astype(np.float32)
user_features=interactions.tocsr().astype(np.float32)
item_features=interactions.tocsr().astype(np.float32)

train, test = random_train_test_split(interactions, test_percentage=0.2, random_state=42)

epochs = 30

# Train model
model = LightFM(loss='bpr', no_components=200, max_sampled=100, learning_schedule='adagrad', item_alpha=1e-6, learning_rate=0.001)

for epoch in range(epochs):
model.fit_partial(train, user_features=user_features, item_features=item_features, num_threads=3, epochs=1)
print(auc_score(model, test, train_interactions = train, user_features=user_features, item_features=item_features, check_intersections=True).mean())`

I check the train and test matrix there is no Null or Nan so I don't understand.
Can you help me on this case.

Start at the auc_score(...) call in the provided Python snippet and reproduce the NaN result with the shown train/test matrices and model settings. Trace the evaluation inputs and consider the issue complete when the cause of the NaN result is identified and a corrective change is documented or validated.

Written by the indexing model from the issue text.