Naver Project (Score Prediction)

Score prediction

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt

tf.__version__

#2.15.0

from google.colab import drive
drive.mount('/content/drive')

Mounted at /content/drive

Setting Data

xy = np.loadtxt('/content/drive/MyDrive/Lecture/양재AI_NLP_basic_to_LLMs/1주차/1_DL_basic/1_TF_examples/exercises/data-01-test-score.csv', delimiter=',', dtype=np.float32)

x_train = xy[:, 0:-1]
y_train = xy[:, [-1]]

print(x_train.shape, y_train.shape)

# (25, 3) (25, 1)

make a dataset

dataset = tf.data.Dataset.from_tensor_slices((x_train, y_train)).batch(len(x_train))

W = tf.Variable(tf.random.normal([3, 1]), name='weight')
b = tf.Variable(tf.random.normal([1]), name='bias')

print(W,b)

#<tf.Variable 'weight:0' shape=(3, 1) dtype=float32, numpy=
#array([[-1.8743812],
#       [-1.1022243],
#       [ 1.3199434]], dtype=float32)> <tf.Variable 'bias:0' shape=(1,) dtype=float32, #numpy=array([-1.426634], dtype=float32)>

def linear_regression(features):
    hypothesis  = tf.matmul(features, W) + b
    return hypothesis

print(linear_regression(x_train))

#tf.Tensor(
#[[-127.43866 ]
# [-149.98508 ]
# [-149.75407 ]
# [-157.39088 ]
# [-118.60724 ]
# [ -78.874275]
# [-110.687904]
# [ -72.05051 ]
# [-132.77861 ]
# [-110.503426]
# [-111.55877 ]
# [-106.60208 ]
# [-160.34053 ]
# [-141.32483 ]
# [-110.140114]
# [-147.12747 ]
# [-140.53905 ]
# [-129.69696 ]
# [-154.26169 ]
# [-137.47736 ]
# [-131.12228 ]
# [-135.53085 ]
# [-126.9178  ]
# [-141.64825 ]
# [-158.47946 ]], shape=(25, 1), dtype=float32)

def loss_fn(hypothesis, labels): #MSE
    cost = tf.reduce_mean(tf.square(hypothesis - labels))
    return cost

optimizer = tf.compat.v1.train.GradientDescentOptimizer(learning_rate=0.00001)

epochs = 5000

for step in range(epochs):
  for features, labels in dataset:
    with tf.GradientTape() as tape:
      loss_value = loss_fn(linear_regression(features),labels)
      grads = tape.gradient(loss_value, [W,b])
      optimizer.apply_gradients(grads_and_vars=zip(grads,[W,b]))
      if step % 100 == 0:
            print("Iter: {}, Loss: {:.4f}".format(step, loss_fn(linear_regression(features),labels)))

#Iter: 0, Loss: 32221.2480
#Iter: 100, Loss: 60.0295
#Iter: 200, Loss: 56.0295
#Iter: 300, Loss: 52.3404
#Iter: 400, Loss: 48.9370
#Iter: 500, Loss: 45.7966
#Iter: 600, Loss: 42.8979
#Iter: 700, Loss: 40.2216
#Iter: 800, Loss: 37.7502
#Iter: 900, Loss: 35.4670
#Iter: 1000, Loss: 33.3574
#Iter: 1100, Loss: 31.4075
#Iter: 1200, Loss: 29.6047
#Iter: 1300, Loss: 27.9373
#Iter: 1400, Loss: 26.3948
#Iter: 1500, Loss: 24.9673
#Iter: 1600, Loss: 23.6459
#Iter: 1700, Loss: 22.4222
#Iter: 1800, Loss: 21.2887
#Iter: 1900, Loss: 20.2383
#Iter: 2000, Loss: 19.2647
#Iter: 2100, Loss: 18.3620
#Iter: 2200, Loss: 17.5246
#Iter: 2300, Loss: 16.7476
#Iter: 2400, Loss: 16.0264
#Iter: 2500, Loss: 15.3567
#Iter: 2600, Loss: 14.7347
#Iter: 2700, Loss: 14.1566
#Iter: 2800, Loss: 13.6193
#Iter: 2900, Loss: 13.1197
#Iter: 3000, Loss: 12.6549
#Iter: 3100, Loss: 12.2224
#Iter: 3200, Loss: 11.8197
#Iter: 3300, Loss: 11.4448
#Iter: 3400, Loss: 11.0954
#Iter: 3500, Loss: 10.7698
#Iter: 3600, Loss: 10.4662
#Iter: 3700, Loss: 10.1831
#Iter: 3800, Loss: 9.9189
#Iter: 3900, Loss: 9.6723
#Iter: 4000, Loss: 9.4421
#Iter: 4100, Loss: 9.2270
#Iter: 4200, Loss: 9.0260
#Iter: 4300, Loss: 8.8382
#Iter: 4400, Loss: 8.6624
#Iter: 4500, Loss: 8.4981
#Iter: 4600, Loss: 8.3442
#Iter: 4700, Loss: 8.2002
#Iter: 4800, Loss: 8.0654
#Iter: 4900, Loss: 7.9390

print(W, b)

#<tf.Variable 'weight:0' shape=(3, 1) dtype=float32, numpy=
#array([[0.07136216],
#       [0.57992035],
#       [1.3724346 ]], dtype=float32)> <tf.Variable 'bias:0' shape=(1,) dtype=float32, #numpy=array([-1.4270566], dtype=float32)>