How can I generate the feature columns to feed to the TabNetClassifier from numpy matrix?

[yxu02]

All the examples used tensorflow out-of-box dataset, which are in tensorflow dataset format.

I'd like to give tabnet a try but I have to use numpy matrix. Even though I converted numpy matrix to tensor via from_tensor_slices, it still didn't work coz I don't have the mapping between my tensor and the feature_columns. Please advise.

[csetraynor]

Hello,

I faced similar issue and I tried the following and it runed, but I'm unsure is correct. I've followed the example and create a dictionary. But from a numpy matrix it needs to be transpose to match tensor dimnesions. It would be great if the devs can have a look and double-check this is correct. Thanks,

col_names = ['a', 'b', 'c'] ## <--- feature colnames here
features_train = tf.cast(X_train, tf.float32)
labels_train = tf.cast(y_train, tf.float32)
x_train = dict(zip(col_names, tf.transpose( features_train )))
dataset_train = tf.data.Dataset.from_tensor_slices((x_train, labels_train)).batch(BATCH_SIZE)