Classification problem using feed forward network

Hello, I faced a problem when I tried to apply feedforward network to classification. This is my code:

net = feedforwardnet(10);

net.performFcn = 'crossentropy';

net.performParam.regularization = 0.1;

net.performParam.normalization = 'none';

[net,tr]=train(net,trainSamples.P,trainSamples.T);

output_norm=sim(net,testSamples.P);

Error occured: 'Performance function replaced with squared error performance.'

It seems the error occured since the performance function for trainlm must be squred error.

But I can still have an output:

The result is ok for me.

I confirmed that the performance function was replaced by mes. So I changed 'net.performFcn = 'crossentropy' to 'net.performFcn = 'mse', and ran the code again. But the result was:

I don't know what happened. I changed the dataset and ran several timee but results were similar. The first condition made a better results for me. Do you have any comment on it？ Thanks