Train a sequential nnet

train_sequential.rs

for epoch in 1..N_EPOCHS {
        let loss = net.forward(&train_data).cross_entropy_for_logits(&train_lbl);
        // backward step 
        opt.backward_step(&loss);
        //accuracy on test
        let val_accuracy = net.forward(&val_data).accuracy_for_logits(&val_lbl);
        println!(
            "epoch: {:4} train loss: {:8.5} val acc: {:5.2}%",
            epoch,
            f64::from(&loss),
            100. * f64::from(&val_accuracy),
        );
    }