ANN - Regression

Code

[Notice] download here

Observing the dataset

    AT	    V	    AP	    RH	    PE
    8.34	40.77	1010.84	90.01	480.48
    23.64	58.49	1011.4	74.2	445.75
    29.74	56.9	1007.15	41.91	438.76
    19.07	49.69	1007.22	76.79	453.09
    11.8	40.66	1017.13	97.2	464.43
    13.97	39.16	1016.05	84.6	470.96
    22.1	71.29	1008.2	75.38	442.35
    ...

AT: Average Temperature

V: Vacuum

AP: Average Pressure

RH: Relative Huminity

PE: Predicted Energy

상기 데이터셋을 활용하여 하기 인공 신경망을 구축하고 새로운 데이터에 대해 효과적으로 예측값을 도출하는 것이 목표이다. Our goal is to build an ANN that refers to the dataset above and make predictions effectively.

Data Preprocessing

Importing the dataset

import numpy as np
import pandas as pd
import tensorflow as tf

dataset = pd.read_excel('Folds5x2_pp.xlsx')
X = dataset.iloc[:, :-1].values
y = dataset.iloc[:, -1].values

Splitting the dataset into the Training set and Test set

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 0)

Building the ANN

Initializing the ANN

ann = tf.keras.models.Sequential()

ann.add(tf.keras.layers.Dense(units=6, activation='relu')) # the input layer and the first hidden layer
ann.add(tf.keras.layers.Dense(units=6, activation='relu')) # the second hidden layer
ann.add(tf.keras.layers.Dense(units=1)) # the ouput layer

Training the ANN

ann.compile(optimizer = 'adam', loss = 'mean_squared_error') # Compiling the ANN
ann.fit(X_train, y_train, batch_size = 32, epochs = 100) # Training the ANN model on the Training set

    ...
    Epoch 12/100
    240/240 [==============================] - 0s 1ms/step - loss: 265.4893
    Epoch 13/100
    ...
    Epoch 99/100
    240/240 [==============================] - 0s 1ms/step - loss: 26.8439
    Epoch 100/100
    240/240 [==============================] - 0s 1ms/step - loss: 26.7392
    <tensorflow.python.keras.callbacks.History at 0x7f895a5850b8>

y_pred = ann.predict(X_test)
np.set_printoptions(precision=2) # precision: decimal places
print(np.concatenate((y_pred.reshape(len(y_pred),1), y_test.reshape(len(y_test),1)),1))

    [[430.79 431.23]
    [461.8  460.01]
    [465.29 461.14]
    ...
    [472.51 473.26]
    [439.39 438.  ]
    [458.55 463.28]]

예측값과 실제값이 수직으로 배열된 2차원 행렬을 통해 가독성 좋게 비교가 가능하다. Through our 2-dimensional matrix, we can easily compare estimates with actual values.