Axon.Loop.run(loop, train_data, %{}, iterations: 1000) Epoch: 0, Batch: 950, loss: 0.0563023 The output batch 950 is a little confused since the input tensor also have batch property.