ConverJens

@dineshvelmuruga @wyljpn I believe that you are both setting your env vars incorrectly. While the values are likely correct, they need to be set at container runtime, not at pipeline...

> > @dineshvelmuruga What do you mean "read the image"? Do you mean read data with ExampleGen from S3? > > If you set the env vars on your pipeline...

@wyljpn So I think that you have done everything correctly up till this point. If I'm reading this correct you are now reading data from your pvc and csv example...

@wyljpn I don't know if there was a missunderstanding but now you are missing all the s3 related beam args that you previously posted. These also need to be present:...

@wyljpn No worries, happy to help :) Great that you go it working! When I started using TFX with Minio 2 years ago there wasn't a single comprehensive post about...

@deep-diver This is because you have only specified an input_fn that can handle your raw data. You need to create an additional one which exposes the transformed data to the...

@deep-diver But then it sounds like you are doing an on-the-fly transformation of your labels in the Trainer? If so, that can definitely cause the issue. But if you have...

I mean that you are missing a serving signature (I think I wrote input_fn before which is an error on my part). The input_fn is just for training so that...

@deep-diver I believe that you are right but about the labels then. But you mention that you have flattened your label into a single array in tfrecord format but you...

@deep-diver And I also feel that your serving signatures should probably look more like this (but I might be wrong): ``` def make_serving_signatures(model, tf_transform_output: tft.TFTransformOutput): model.tft_layer = tf_transform_output.transform_features_layer() @tf.function(input_signature=[ tf.TensorSpec(shape=[None],...