我参加了udemy课程＆＃34;深度学习A-Z™：动手人工神经网络＆＃34;在第一个作业（构建人工神经网络）中，我运行以下代码
# Import the libraries
# Opensource computation numbers GPU enabled
# import Theano as th
# import Tensorflow as tf
# import Keras as ke
# Part 1 Data preprocessing
# Importing the libraries
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
# Get the data set
dataset = pd.read_csv('Churn_Modelling.csv')
# Pull test sample
testset = pd.read_csv('Churn_Modelling_Test.csv')
# Dividing the dependent and independent variable
# -1 expresses the last column (feature)
x = dataset.iloc[:, 3:13].values
y = dataset.iloc[:, 13].values
# Get the independent variables
testx = testset.iloc[:, 3:13].values
# Pre processing
# Concatenate the test rows
x = np.vstack((x,testx))
# Taking care of categorical variables
# Creating encoding for categorical data
from sklearn.preprocessing import LabelEncoder, OneHotEncoder
labelencoder_x_country = LabelEncoder()
x[:,1] = labelencoder_x_country.fit_transform(x[:,1])
labelencoder_x_gender = LabelEncoder()
x[:,2] = labelencoder_x_gender.fit_transform(x[:,2])
# Block the ANN trying to assert true value of encoding
# Create dummy variables for the non-dichotomic encoded variable
onehotencoder = OneHotEncoder(categorical_features = [1])
x = onehotencoder.fit_transform(x).toarray()
# Take care of the dummy variable in order not to have more than 2
x = x[:,1:]
prepro_testx = np.array([x[1000,:]])
x = np.delete(x, 10000, 0)
# 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)
# Scaling the features (columns) to standardize or normalize values
from sklearn.preprocessing import StandardScaler
sc_x = StandardScaler()
x_train = sc_x.fit_transform(x_train)
x_test = sc_x.transform(x_test)
# Scale it
prepro_testx = sc_x.transform(prepro_testx)
# Part 2 Making the ANN
# Importing the libraries
import keras as ke
# Start de ANN
from keras.models import Sequential
# Make the layers
from keras.layers import Dense
# Initialize the ANN you can define a graph or a sequence of layers
# Create classifier
# Classifier is the model of the neural network
classifier = Sequential()
# Construct the layers with the input layer and first hidden layer
# The number of nodes will be the number of independent variables
# For the hidden layers, we will use the rectifier function
# For the outer layers, we will use the sigmoid function
# The output is 0 or 1 so only one node
aux = ((x[1].size + 1)/2)
# Api 2
classifier.add(Dense(activation = 'relu', input_dim = x[1].size, units = int(aux),
kernel_initializer='uniform'))
# Api 1
#classifier.add(Dense(output_dim = int(aux), init = 'uniform',
# activation = 'relu', input_dim = x[1].size))
# Add the second hidden layer
classifier.add(Dense(activation = 'relu', units = int(aux),
kernel_initializer='uniform'))
# Add the output layer
classifier.add(Dense(activation = 'sigmoid', units = 1,
kernel_initializer='uniform'))
# Compiling the ANN (Applying SGD to the optimizer with adam mode)
# SGD = optimizer, Loss function = SUM(y-y`)^2->min but not in this case
# The activation is a sigmoid and not a linear regression, we use Log loss
# For two output categories use binary_crossentropy else categorical_...
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy',
metrics = ['accuracy'])
# Fitting the classifier (ANN) to the Training Set
classifier.fit(x_train, y_train, batch_size = 10, epochs = 100)
# Predicting the Test set results
y_pred = classifier.predict(x_test)
y_pred = (y_pred > 0.5)
new_pred = classifier.predict(prepro_testx)
new_pred = (new_pred > 0.5)
# Making the Confusion Matrix
from sklearn.metrics import confusion_matrix
cm = confusion_matrix(y_test, y_pred)
对于预测集合，它会运行并抛出一个False。但是，在本教程中，他们不会从额外的csv中提取，他们只是将硬编码值放入
new_pred = classifier.predict(sc_x.transform(np.array([[1, 0, 416, 0, 41, 10, 122189.66,2,1,0,98301.61]])))
new_pred = (new_pred > .50)
两个代码都会运行并返回false，但sc_x.transform(np.array([[1, 0, 416, 0, 41, 10, 122189.66,2,1,0,98301.61]]))和prepro_testx的缩放值会有所不同。
为什么会这样？
我的不同之处是：
添加新的CSV，以防我想添加几个额外的样本而不只是一个。 将新CSV附加到主采样。 对所有数据进行编码/缩放，以便以相同的方式完成。 在拆分之前检索附加的额外样本。 分类
他们做的是：
使用sc_x
有些值相似但有些则不相似。 哪一个是正确的？