Wsrtjtyk

I am trying to build a CNN network with tensorflow package. My train data sets are 237 images with 100X100 in 6 classes and my test sets are 30 images with same size and classes.
Everything is going well except last part that I want to measure accuracy with

acc=tf.reduce_mean(tf.cast(matches,tf.float32))

I cannot understand why it make error while I am testing with test sets (both data and label) that it is compatible “(30,100,100,1) vs (30,6)”.
Can anyone explain why and what is possible solution.

Note: when I apply batch size 30 (equal to number my test set row) on train data it works.i assume the model expect that the train and test set should be equal number of row that it doesn't make set (usually 70% train vs 30% test) i checked some post with same error title but i couldn’t get it, in face, i think although the error title was same but their problem was different. mostly they had issue with flatten layer.

screenshot from data set info
enter image description here

here is the simple code

…

convo_1=…(x_img,shape=[5,5,1,32])

convo_1_pooling=…(convo_1)



convo_2=…(x_img,shape=[5,5,32,64])

convo_2_pooling=…(convo_2)



convo_3=…(x_img,shape=[5,5,64,128])

convo_3_pooling=…(convo_3)



convo_3_flat=tf.reshape(convo_3_pooling,[-1,13*13*128])

full_layer_one=tf.nn.relu(sp.normal_full_layer(convo_3_flat,21632))



# DROPOUT

hold_prob=tf.placeholder(tf.float32)

full_one_dropout=tf.nn.dropout(full_layer_one,keep_prob=hold_prob)

y_pred=sp.normal_full_layer(full_one_dropout,6)



# LOOS FUNCTION

cross_entropy=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y,logits=y_pred))



# OPTIMIZER

optimizer=tf.train.AdamOptimizer(learning_rate=learning_rate)

train=optimizer.minimize(cross_entropy)



with tf.Session() as sess:

    sess.run(init)

    b=0

    for i in range(training_iters):

        sess.run(train,feed_dict={x_img:train_x, y:train_y,hold_prob:.5})

        if i%2 ==0:

            print("NO STEP:{}",format(i))

            print("ACCURACY: ")

            matches=tf.equal(tf.argmax(y_pred,1),tf.argmax(train_y,1))

            acc=tf.reduce_mean(tf.cast(matches,tf.float32))

            print(sess.run(acc,feed_dict={x_img:test_x, y:test_y, hold_prob:1.0}))

            print('n')

and this is the error

InvalidArgumentError: Incompatible shapes: [30] vs. [237]

     [[{{node Equal_1}} = Equal[T=DT_INT64, _device="/job:localhost/replica:0/task:0/device:CPU:0"](ArgMax_2, ArgMax_3)]]

thank you so much