How to get ROC_Curve and Confusion Matrix in TensorFlow

Question

I am trying to get an roc_curve and confusion matrix in Tensorflow. I used the sklearn.metrics function and I am getting an error. My code is below: from sklearn.metrics import roc_curve, auc

n_inputs = x_train.shape[1]
n_hidden1 = 500
n_hidden2 = 200
n_outputs = 2
learning_rate = 0.01

X = tf.placeholder(tf.float32, shape=(None, n_inputs), name="X")
y = tf.placeholder(tf.int64, shape=(None), name="y")

hidden1 = tf.layers.dense(X, n_hidden1, activation=None)
hidden2 = tf.layers.dense(hidden1, n_hidden2, activation=None)
logits = tf.layers.dense(hidden2, n_outputs)

loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y, logits=logits))

training_op = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss)

correct = tf.nn.in_top_k(logits, y, 1)
accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))

init = tf.global_variables_initializer()

n_epochs = 20

with tf.Session() as sess:
    init.run()
    for epoch in range(n_epochs):
        sess.run(training_op, feed_dict={X: x_train, y: y_train})
        acc_train = accuracy.eval(feed_dict={X: x_train, y: y_train})
    acc_test = accuracy.eval(feed_dict={X: x_test, y: y_test})
    print("Epoch:", epoch, "Train accuracy:", acc_train, "Test accuracy:", acc_test)

    y_score = np.array(logits)
    roc_curve(y_test, y_score)

The error I got is below:

TypeError: Singleton array array(<tf.Tensor 'dense_26/BiasAdd:0' shape=(?, 2) dtype=float32>, dtype=object) cannot be considered a valid collection.

Any help will be appreciated. Thanks!

Answer 1

When you call

sess.run(training_op, feed_dict={X: x_train, y: y_train})

You need to request that the network return the value of the logits tensor, change it to this:

training_op_result, logits_result = sess.run([training_op, logits], feed_dict={X: x_train, y: y_train})
y_score = np.array(logits_result)
roc_curve(y_test, y_score)

A tensor is a graph object. You access value/results of a tensor or computation through sess.run.