1-2. Artificial Nuerons 구현 (1)

 

 

 

 

Affine Function¶

 

 

 

 

In [2]:

import tensorflow as tf

from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import Conv2D

x = tf.constant([[10.0]])  # input setting (Note : input은 matrix 형태 )
dense = Dense(units=1, activation="linear")  # imp.an affine function

y_tf = dense(x)  # forward propagation + p arams initialization
W, B = dense.get_weights()  # get weight, bias

# print results
print("========Input/Weight/Bias=========")
print("x : {}\n{}\n".format(x.shape, x.numpy()))
print("W : {}\n{}\n".format(x.shape, W))
print("B : {}\n{}\n".format(x.shape, B))

y_man = tf.linalg.matmul(x, W) + B  # forward propagation(manual) 행렬의 곱으로 증명

print("========Output/Weight/Bias=========")
print("y(Tensorflow):{}\n{}\n)".format(y_tf.shape, y_tf.numpy()))
print("y(Manual): {}\n{}\n".format(y_man.shape, y_man.numpy()))

 

 

========Input/Weight/Bias=========
x : (1, 1)
[[10.]]

W : (1, 1)
[[-0.7190732]]

B : (1, 1)
[0.]

========Output/Weight/Bias=========
y(Tensorflow):(1, 1)
[[-7.190732]]
)
y(Manual): (1, 1)
[[-7.190732]]

 

 

결과가 같음을 알수있음¶

 

 

Params Initalization¶

 

In [1]:

import tensorflow as tf

from tensorflow.keras.layers import Dense
from tensorflow.keras.initializers import Constant

x = tf.constant([[10.]])  # input setting(Note : input -> matrix)


# weingt/bias setting
w, b = tf.constant(10.), tf.constant(20.) # float로 쓰는걸 습관으로 하기
w_init, b_init = Constant(w), Constant(b)
# target W, B설

dense = Dense(units=1,
              activation='linear',
              kernel_initializer=w_init,
              bias_initializer=b_init) 

# affine만들고 activation지정 , init으로 초기화 지정

y_tf = dense(x)

print(y_tf)

w, b = dense.get_weights()

print("W :{}\n{}\n".format(w.shape, w))

print("B :{}\n{}\n".format(b.shape, b))

 

 

tf.Tensor([[120.]], shape=(1, 1), dtype=float32)
W :(1, 1)
[[10.]]

B :(1,)
[20.]

 

 

10*10 +20 = 120 이 결과로 나온거

 

 

AFfine Function with n Features¶

 

In [4]:

import tensorflow as tf

from tensorflow.keras.layers import Dense

x = tf.random.uniform(shape=(1, 10), minval=0, maxval=10) # 0~10사이의 숫자

dense = Dense(units=1)

y_tf = dense(x) # Weight 가 10개로 적용

W, B = dense.get_weights()
# 알아서 초기화 됨

y_man = tf.linalg.matmul(x, W)+B

print('========Input/Weight/Bias=========')
print("x : {}\n{}\n".format(x.shape, x.numpy()))
print("W : {}\n{}\n".format(W.shape, W))
print("B : {}\n{}\n".format(B.shape, B))

print('========output/Weight/Bias=========')
print("y(Tensorflow):{}\n{}\n)".format(y_tf.shape, y_tf.numpy()))
print("y(Manual): {}\n{}\n".format(y_man.shape, y_man.numpy()))

 

 

========Input/Weight/Bias=========
x : (1, 10)
[[7.19821   2.8476775 4.774387  7.7870045 9.054894  3.8097656 6.4105678
  2.5757134 2.0627534 5.32243  ]]

W : (10, 1)
[[ 0.46323365]
 [ 0.06407285]
 [ 0.7176154 ]
 [ 0.50136095]
 [-0.707779  ]
 [ 0.10467267]
 [-0.21442622]
 [ 0.3111828 ]
 [-0.21710354]
 [-0.0219177 ]]

B : (1,)
[0.]

========output/Weight/Bias=========
y(Tensorflow):(1, 1)
[[3.6995368]]
)
y(Manual): (1, 1)
[[3.6995368]]

 

In [ ]: