Jiayi Zhao

TFX has its own cache, if all input artifact & exec properties (in this case, input is just the query) are the same, it will use last component execution's result....

Hi, we have a [example](https://github.com/tensorflow/tfx/blob/master/tfx/examples/chicago_taxi_pipeline/taxi_utils.py#L122) for vocabulary in TFT

It would be a good candidate for TFX-[addons](https://github.com/tensorflow/tfx-addons)

The API doc is [here](https://www.tensorflow.org/tfx/api_docs/python/tfx/v1/proto) but since it's proto file, the easiest way is checking the [definition](https://github.com/tensorflow/tfx/blob/master/tfx/proto/example_gen.proto) itself, which has comments for the data structure

Hi, Kamal, currently we support once shot pipeline which can be scheduled daily with different input path as different data window, long term we have plan for continuous training with...

currently there are two ways to use one-shot batch for multiple data span 1) prepare the data span in different paths, e.g., /tmp/YYYY-MM-DD/data, trigger the pipeline with a different [input_config...

oh, I see what you mean by windowing data here are two separate problems 1) from raw data to tf examples (or sequence examples that we don't support yet) 2)...

let's keep this as a feature request

we don't support streaming data processing. We support t[raining based on rolling window of spans](https://github.com/tensorflow/community/blob/master/rfcs/20210115-tfx-periodic-training.md). But I guess you want to sessionize some raw logs into training examples, consider using...

Do you want your model being updated streamingly? or you want to process your log streaming? log processing pipeline can be separate with training pipeline, while training part takes training...