jameswex
commented
3 years ago WIT doesn't have great support for visualizing sequence data itself, but I believe you should be able to use WIT for your model and dataset. I would suggest trying it in notebook mode and using a custom prediction function you write yourself that calls your TF model directly for doing predictions.
Related tutorials: https://pair-code.github.io/what-if-tool/learn/tutorials/notebooks/ https://pair-code.github.io/what-if-tool/learn/tutorials/custom-prediction/
Let me know if you give it a shot and any issues you run into.
We're training a tensorflow model for time series forecasting. I've seen that it is possible to use SequenceExamples in WIT. We would like to use an approach which generates the windowing of the data as part of the model graph. We're using tensorflow functions for windowing the data - we're experimenting with tf.keras.preprocessing.timeseries_dataset_from_array and tf.data.Dataset.window-method - and then use this data for training, for example LSTM. Would it be possible to load such a model and use it to generate predictions in WIT?