NSL for multiclass classification

TensorFlow General Discussion
1

Hi,
I am working on multiclass classification using neural structured learning. I am trying to edit the code (for binary classification) for Graph regularization for sentiment classification using synthesized graphs  |  Neural Structured Learning  |  TensorFlow for my purpose. I am getting the following error. Can anyone please help?

File “/home/vinay/anaconda3/envs/newenvt/lib/python3.9/site-packages/tensorflow/python/eager/execute.py”, line 59, in quick_execute
tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
tensorflow.python.framework.errors_impl.InvalidArgumentError: Feature: NL_nbr_0_words (data type: float) is required but could not be found.
[[{{node ParseSingleExample/ParseExample/ParseExampleV2}}]]
[[IteratorGetNext]] [Op:__inference_train_function_1996]

I am getting datatype errors when I use one hot encoding for labels . So I switched to sparse categorical crossentropy. Am I doing something wrong? What else can be done for multiclass classification?

3

Hi @Vinay_Gupta,

Sorry for the delayed response. You might have had a solution by this time. Here are some suggestions. The InvalidArgument error might be due to the feature NL_nbr_0_words missing in your input data with a data type of float and make sure that the feature available in input data and correctly named. The label encoding should match with the type of Loss function as well. For further assistance, please provide reproducible code .
Thank You.