张量流模型制作器

Question

I'm making an mobile app that uses TensorFlow Lite to perform text classification on tweets. I've done this successfully with the TensorFlow sample model but this model is trained on IMDB movie reviews and I want to have a custom model on device that is trained on tweets to increase accuracy. I have training and test sets for this domain and am trying to create a custom model following this tutorial https://www.tensorflow.org/lite/tutorials/model_maker_text_classification

I'm running into a pythion KeyError though and cant figure out why. here is a screen shot

You can see on the right a sample of my csv, I have a label and Sentence column, using the TextClassifierDataLoader. I don't understand why this key error is happening, I'm interpreting this as it can't find a column named "Sentence" but cleary it's there.

Any ideas?

Answer 1

Seems to me that is_training is not an available argument in the class, so just take out that line/argument if you want it to compile (what it was supposed to do in the first place, I haven't looked into).

https://github.com/tensorflow/examples/blob/1dc6978e2141e7a5efebcf6971b3afa9cb055679/tensorflow_examples/lite/model_maker/core/data_util/text_dataloader.py#L90

Answer 2

问题是“句子”列的开头有一个空格