Remove values from a tensor that are within a given range

user_5113410 0
Viewed 1268

I want to be able to know the rounded accuracy of my neural network when the prediction is above or below a certain threshold. For example, I want it to only calculate accuracy when the prediction is above 0.55 or below 0.45 in order to filter out near 50/50 cases.

I tried using the soft_acc function on stackoverflow and adding an if else to the beginning to filter out the near 50/50s.

def soft_acc(y_true, y_pred):
    if y_pred > 0.55 or y_pred < 0.45:
        return K.mean(K.equal(K.round(y_true), K.round(y_pred)))

I received the following error message.

TypeError: Using a tf.Tensor as a Python bool is not allowed. Use if t is not None: instead of if t: to test if a tensor is defined, and use TensorFlow ops such as tf.cond to execute subgraphs conditioned on the value of a tensor.

tensor tensorflow python
1 Answers

Use tf.boolean_mask to filter out values at indices that don't meet the required threshold.

# remove values from `X` in interval (lo, hi)
mask = tf.math.logical_or(tf.lesser(X, lo), tf.greater(X, hi))
X = tf.boolean_mask(X, mask)

In your case, you would define soft_acc as

def soft_acc(y_true, y_pred):
    mask = tf.math.logical_or(tf.greater(y_pred, 0.55), tf.lesser(y_pred, 0.45))
    y_true2 = tf.boolean_mask(y_true, mask)
    y_pred2 = tf.boolean_mask(y_pred, mask)

    return K.mean(K.equal(K.round(y_true2), K.round(y_pred2)))
user_49090870
Related