Restore and Predict in Tensorflow

Question

I am following this link for cancer prediction. Now after my training & test phase is done i want to give new data as input and want the prediction. For that i am saving the model and restoring it to get the prediction, but i am getting error as

ValueError: Cannot feed value of shape (31,) for Tensor 'Placeholder_1:0', which has shape '(?, 31)'

Below is my code:

saver = tf.train.Saver()
sampletest = [-0.24222039 -0.75688274 -0.26264569 -0.75637054 -0.7154845  -0.55675554 -0.51883267 -0.69442359 -0.87362527 -1.46135011 -0.05206671 -0.2790065 -0.28614862 -0.1934161  -0.38264881 -0.1295509   0.05817795 -0.32080093-0.64650773 -0.19383338 -0.14508449 -0.74260509 -0.66173979 -0.73123076-0.68635871 -0.78697688 -0.4790055  -0.71702336 -0.90543288 -1.1197415-0.41889736]
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    for epoch in range(training_epochs):
        for batch in range(int(n_samples / batch_size)):
            batch_x = input_X[batch * batch_size: (1 + batch) * batch_size]
            batch_y = input_Y[batch * batch_size: (1 + batch) * batch_size]
            print(batch_x[0])

            sess.run([optimizer], feed_dict={x: batch_x,
                                             y_: batch_y,
                                             pkeep: training_dropout})
            saver.save(sess,'.\cancer_model')

        # Display logs after every 10 epochs
        if (epoch) % display_step == 0:
            train_accuracy, newCost = sess.run([accuracy, cost],
                                               feed_dict={x: input_X, y_: input_Y, pkeep: training_dropout})

            valid_accuracy, valid_newCost = sess.run([accuracy, cost],
                                                     feed_dict={x: input_X_valid, y_: input_Y_valid, pkeep: 1})

            print("Epoch:", epoch, "Acc =", "{:.5f}".format(train_accuracy), "Cost =", "{:.5f}".format(newCost),
                  "Valid_Acc =", "{:.5f}".format(valid_accuracy), "Valid_Cost = ", "{:.5f}".format(valid_newCost))

            # Record the results of the model
            accuracy_history.append(train_accuracy)
            cost_history.append(newCost)
            valid_accuracy_history.append(valid_accuracy)
            valid_cost_history.append(valid_newCost)

            # If the model does not improve after 15 logs, stop the training.
            if valid_accuracy < max(valid_accuracy_history) and epoch > 100:
                stop_early += 1
                if stop_early == 15:
                    break
            else:
                stop_early = 0

    print("Optimization Finished!")

with tf.Session() as sess:
    saver = tf.train.import_meta_graph('.\Cancer_Model\cancer_model.meta')
    saver.restore(sess, tf.train.latest_checkpoint('.\Cancer_Model'))
    prediction = sess.run(y4,feed_dict={x:sampletest})
    print(prediction)

Please help me with this.

Answer 1

The problem is that you model expects a batch of examples, and you are just giving one. Try replacing:

prediction = sess.run(y4, feed_dict={x: sampletest})

With:

prediction = sess.run(y4, feed_dict={x: [sampletest]})

Then you will get a "batch" of results in prediction with a single element.

Answer 2

i guess since the model is restored,the placeholder of input of y4 was renamed to Variable_1 to avoid naming confusing of graph variable, try this and see

prediction = sess.run(y4,feed_dict={"Variable_1:0":[sampletest]})

though i think prediction = sess.run(y4,feed_dict={"Variable_1:0":sampletest}) would work too and also you should restore y4 as

y_4 = graph.get_operation_by_name('y4:0')

then run

prediction = sess.run(y_4,feed_dict={"Variable_1:0":[sampletest]})