minst_nn_tf.py

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import begin

l1_nodes = 200
l2_nodes = 100
final_layer_nodes = 10

# define placeholder for data
# also considered as the "visibale layer, the layer that we see"
X = tf.placeholder(dtype=tf.float32, shape=[None, 784])
# placeholder for correct labels
Y_ = tf.placeholder(dtype=tf.float32)
# define weights / layers here
# needs weights and bias for each layer in the network. Input to one layer is the
# output from the previous layer
w1 = tf.Variable(initial_value=tf.truncated_normal([784, l1_nodes], stddev=0.1))
b1 = tf.Variable(initial_value=tf.zeros([l1_nodes]))
Y1 = tf.nn.relu(tf.matmul(X, w1) + b1)

w2 = tf.Variable(initial_value=tf.truncated_normal([l1_nodes, l2_nodes], stddev=0.1))
b2 = tf.Variable(tf.zeros([l2_nodes]))
Y2 = tf.nn.relu(tf.matmul(Y1, w2) + b2)

w3 = tf.Variable(initial_value=tf.truncated_normal([l2_nodes, final_layer_nodes], stddev=0.1))
b3 = tf.Variable(tf.zeros([final_layer_nodes]))
Y = tf.nn.softmax(tf.matmul(Y2, w3) + b3)

# defien cost function and evaluation metric
cross_entropy = -tf.reduce_sum(Y_ * tf.log(Y))
is_correct = tf.equal(tf.argmax(Y, 1), tf.argmax(Y_, 1))
accuracy = tf.reduce_mean(tf.cast(is_correct, tf.float32))

# gradient descent
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.003)
train_step = optimizer.minimize(loss=cross_entropy)

init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)

@begin.start
def main(n_iter : "number of iterations in model"):

    mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

    for i in range(int(n_iter)):
        batch_X, batch_y = mnist.train.next_batch(100)
        # link previously defined placeholders to incoming data
        train_data = {X: batch_X, Y_: batch_y}

        # train
        sess.run(train_step, feed_dict=train_data)
        # training accuracy and cost
        train_a, train_c = sess.run([accuracy, cross_entropy], feed_dict=train_data)
        # test accuract and cost
        test_data = {X: mnist.test.images, Y_: mnist.test.labels}
        test_a, test_c = sess.run([accuracy, cross_entropy], feed_dict=test_data)

        if i % 100 == 0:
            print("Train accuracy: {}, Test accuracy: {}".format(train_a, test_a))

When I try running the above script I get the following error:
File "anuram.py", line 3, in
import begin
ModuleNotFoundError: No module named 'begin'

ah, just remove the "import begin" line from the top of the script

or rather pip install begins if you want to run this script from the command line