How do you decode one-hot labels in Tensorflow?

python tensorflow machine-learning deep-learning one-hot-encoding
Matt Camp picture Matt Camp · Dec 30, 2016 · Viewed 20k times · Source

Been looking, but can't seem to find any examples of how to decode or convert back to a single integer from a one-hot value in TensorFlow.

I used tf.one_hot and was able to train my model but am a bit confused on how to make sense of the label after my classification. My data is being fed in via a TFRecords file that I created. I thought about storing a text label in the file but wasn't able to get it to work. It appeared as if TFRecords couldn't store text string or maybe I was mistaken.

Answer

martianwars picture martianwars · Dec 30, 2016

You can find out the index of the largest element in the matrix using tf.argmax. Since your one hot vector will be one dimensional and will have just one 1 and other 0s, This will work assuming you are dealing with a single vector.

index = tf.argmax(one_hot_vector, axis=0)

For the more standard matrix of batch_size * num_classes, use axis=1 to get a result of size batch_size * 1.

Related questions

How to apply gradient clipping in TensorFlow?

Considering the example code. I would like to know How to apply gradient clipping on this network on the RNN where there is a possibility of exploding gradients. tf.clip_by_value(t, clip_value_min, clip_value_max, name=…

Read More
Unbalanced data and weighted cross entropy

I'm trying to train a network with a unbalanced data. I have A (198 samples), B (436 samples), C (710 samples), D (272 samples) and I have read about the "weighted_cross_entropy_with_logits" but all the examples I found are for binary …

Read More
How does keras define "accuracy" and "loss"?

I can't find how Keras defines "accuracy" and "loss". I know I can specify different metrics (e.g. mse, cross entropy) - but keras prints out a standard "accuracy". How is that defined? Likewise for loss: I know I can …

Read More