# Tensorflow : Sequence when writing tensorflow

1. Write input matrix 

• It is basically x and y matrix (x = [1,2,3], y=[4,5,6])

2. Write placeholder 

• Placeholder would be filled by matrix, so we gotta make x and y placeholder
• x = tf.placeholder(tf.float32) <- make float type placeholder to get a value from matrix x

3. Write variables

• Variables would be filled basically by random seed
• So, we gotta make variables for weight(w) and b(bias)

4. Write hypothesis

• It is differed from case by case
• For example: hypothesis would be "x1*w1 + x2*w2 + .... + b"

5. Write cost function

• Use 'tf.reduce_mean' function with calculation for cost function
• It would be: cost = tf.reduce_mean(hypothesis - y)

6. Write optimizer

• This sequence is written for learning rate 
• Learning rate is important to get a minimum cost value with a short time
• Use tf.train.GradientDescentOptimizer with learning rate
• For example: optimizer = tf.train.GradientDescentOptimizer(learning_rate=x)
• x is differed from case by case

7. Write training set

• Users can minimize cost value with learning rate using this function
• For example: train = optimizer.minimize(cost)

8. Initialize Session

• Users should initialize Session while using Tensorflow
• Simply use 'sess = tf.Session()'

9. Initialize Variables

• Variables should be initialized before they are used
• Initialization must be run by sess.run()
• For example: sess.run(tf.global.variables_initializer())

10. Calculation

• Generally, Calculating in Tensorflow is run by recursive function
• for step in range('how many times user want'):
• Using 'cost value', 'hypothesis value' and 'train value'
• Users should substitute these values using sess.run also
• In python, users can substitue multiple placeholders one time
• For example: sess.run([cost, hypothesis, train], feed_dict={x1: matrix_x1, x2: matrix_x2 ... y: matrix_y})