Data Collection
This dataset contains 23 columns and 1456714 records. The dataset reflects reported incidents of crime (with the exception of murders where data exists for each victim) that occurred in the City of Chicago from 2012 to 2017, minus the most recent seven days. Data is extracted from the Chicago Police Department's CLEAR (Citizen Law Enforcement Analysis and Reporting) system. In order to protect the privacy of crime victims, addresses are shown at the block level only and specific locations are not identified. This data includes unverified reports supplied to the Police Department. The preliminary crime classifications may be changed at a later date based upon additional investigation and there is always the possibility of mechanical or human error. Therefore, the Chicago Police Department does not guarantee (either expressed or implied) the accuracy, completeness, timeliness, or correct sequencing of the information and the information should not be used for comparison purposes over time.
Data PreProcessing :
Data preprocessing in Machine Learning is a crucial step that helps enhance the quality of data to promote the extraction of meaningful insights from the data. Data preprocessing in Machine Learning refers to the technique of preparing (cleaning and organizing) the raw data to make it suitable for building and training Machine Learning models.
The data have many missing values. To handle this part, data cleaning is done. It involves handling missing data, noisy data etc. we have converted the Arrest column into a numerical column for implementing the neural network.
Feature selection :
All machine learning algorithms use some input data to create outputs. This input data comprise features, which are usually in the form of structured columns. Algorithms require features with some specific characteristic to work properly.
In the dataset we have considered all the numerical columns as input columns which is also known as Features column. Unnamed:0, ID, Beat, Ward, District, Community, X-Coordinate, Y-Coordinate, Latitude, Longitude These are numerical columns considered as features columns.
Model Selection
The objective of model selection is to find the network architecture with the best generalization properties, that is, that which minimizes the error on the selected instances of the data set (the selection error). We applied two models on the dataset ANN and delta bar delta algorithm and we used an optimizer adadelta in the ANN model.
Adadelta Optimizer : Adadelta optimization is a stochastic gradient descent method that is based on adaptive learning rate per dimension to address two drawbacks:
The continual decay of learning rates throughout training The need for a manually selected global learning rate
Delta bar Delta :
The DBD algorithm is a heuristic approach to improve the convergence speed of the weight in ANNs. The weights are updated by w(k+1)=w(k)+α(k)δ(k). Where α(k) is the learning coefficient and assigned to each connection, δ(k)is the gradient component of the weight change. δ(k) is employed to implement the heuristic for incrementing and decrementing the learning coefficients for each connection.
Model evaluate :
Model evaluation aims to estimate the generalization accuracy of a model in future.
ANN model Accuracy is 0.7400 optimizer - adadelta
ANN model Accuracy is 0.7400 optimizer - adam
Code Implementation
Import necessary packages
## import libraries
import pandas as pd
import numpy as np
location = '/content/drive/My Drive/Chicago_Project/data/Chicago_Crimes.csv'
from google.colab import drive
drive.mount('/content/drive')
# file Location
file_loc =location
# Read file
df = pd.read_csv(file_loc,header=0, sep=',', quotechar='"')
df.head()
Output:
cols = df.select_dtypes([np.number]).columns
cols
output:
Index(['Unnamed: 0', 'ID', 'Beat', 'District', 'Ward', 'Community Area',
'X Coordinate', 'Y Coordinate', 'Year', 'Latitude', 'Longitude'],
dtype='object')
features = df[[i for i in cols]]
features['Arrest'] = df['Arrest']
features.dropna(inplace=True)
features.head()
output:
from sklearn.preprocessing import LabelEncoder
X = features.drop(['Arrest'],axis=1)
y = features['Arrest']
le = LabelEncoder()
y = le.fit_transform(y)
X.head()
output:
# Splitting the dataset into the Training set and Test set
from sklearn.model_selection import train_test_split#.asarray(x).astype('float32')
X_train, X_test, y_train, y_test = train_test_split(np.array(X).astype('float32'), y.astype('float32'), test_size = 0.2, random_state = 42)
# Importing the Keras libraries and packages
from keras.models import Sequential
from keras.layers import Dense
# Initialising
classifier = Sequential()
# Adding the input layer and the first hidden layer
classifier.add(Dense(units=100,activation="relu", input_dim=X_train.shape[1]))
#classifier.add_weight(shape=(5,72))
# Adding the second hidden layer
classifier.add(Dense(250, activation = 'relu'))
# Adding the third hidden layer
classifier.add(Dense(150, activation = 'relu'))
classifier.add(Dense(50, activation = 'relu'))
# Adding the output layer
classifier.add(Dense(units = 1, activation = 'sigmoid'))
# Compiling the delta bar delta algorithim
classifier.compile(optimizer = 'adadelta',
loss = 'mse',
metrics = ['accuracy'])
classifier.fit(X_train, y_train, batch_size = 100, epochs= 5)
Output:
Epoch 1/5
11357/11357 [==============================] - 28s 2ms/step - loss: 0.2620 - accuracy: 0.7380
Epoch 2/5 11357/11357 [==============================] - 25s 2ms/step - loss: 0.2620 - accuracy: 0.7380
Epoch 3/5 11357/11357 [==============================] - 25s 2ms/step - loss: 0.2619 - accuracy: 0.7381
Epoch 4/5 11357/11357 [==============================] - 25s 2ms/step - loss: 0.2615 - accuracy: 0.7385
Epoch 5/5 11357/11357 [==============================] - 24s 2ms/step - loss: 0.2619 - accuracy: 0.7381
classifier.summary()
Output:
## evaluating the model
print(classifier.evaluate(X_test,y_test))
Output:
8873/8873 [==============================] - 13s 1ms/step - loss: 0.2600 - accuracy: 0.7400 [0.2599650025367737, 0.7400349974632263]
# Initialising the ANN
ann = Sequential()
# Adding the input layer and the first hidden layer
ann.add(Dense(units=100, input_dim=X_train.shape[1]))
#classifier.add_weight(shape=(5,72))
# Adding the second hidden layer
ann.add(Dense(250, activation = 'relu'))
# Adding the third hidden layer
ann.add(Dense(150, activation = 'relu'))
ann.add(Dense(50, activation = 'relu'))
# Adding the output layer
ann.add(Dense(units = 1, activation = 'sigmoid'))
# Compiling the delta bar delta algorithim
ann.compile(optimizer = 'adam',
loss = 'mse',
metrics = ['accuracy'])
# Fitting the ANN to the Training set
ann.fit(X_train, y_train, batch_size = 100, epochs= 5)
output:
Epoch 1/5 11357/11357 [==============================] - 25s 2ms/step - loss: 0.2617 - accuracy: 0.7383
Epoch 2/5 11357/11357 [==============================] - 25s 2ms/step - loss: 0.2622 - accuracy: 0.7378
Epoch 3/5 11357/11357 [==============================] - 25s 2ms/step - loss: 0.2616 - accuracy: 0.7384
Epoch 4/5 11357/11357 [==============================] - 25s 2ms/step - loss: 0.2617 - accuracy: 0.7383
Epoch 5/5 11357/11357 [==============================] - 25s 2ms/step - loss: 0.2616 - accuracy: 0.7384
ann.evaluate(X_test,y_test)
Output:
8873/8873 [==============================] - 13s 1ms/step - loss: 0.2600 - accuracy: 0.7400
[0.2599650025367737, 0.7400349974632263]
a = np.array(X)
na = a.reshape(a.shape[0],11,-1)
na=na.reshape(-1,1)
na = na[:20000]
na =na.reshape(-1,20,1)
na.shape
Output:
(1000, 20, 1)
DELTA BAR DELTA
import numpy as np
import matplotlib
import matplotlib.pyplot as plt
num_examples = len(na)
input_dim = 20
meta_step_size = 0.05
h = np.zeros((input_dim, 1))
w = np.random.normal(0.0, 1.0, size=(input_dim, 1))
beta = np.ones((input_dim, 1)) * np.log(0.05)
alpha = np.exp(beta)
s = np.zeros((input_dim, 1))
target_num = 5
def generate_task():
for i in range(target_num):
s[i] = np.random.choice([-1, 1])
def set_target(x, examples_seen):
if examples_seen % 20 == 0:
s[np.random.choice(np.arange(target_num))] *= -1
for i in range(target_num, input_dim):
s[i] = np.random.random()
return np.dot(s.transpose(), x)[0, 0]
def main(X):
alpha_mat = np.zeros((num_examples, 2))
global h, w, beta, alpha
generate_task()
for example_i,x in enumerate(X):
y = set_target(x, example_i + 1)
estimate = np.dot(w.transpose(), x)[0, 0]
delta = y - estimate
beta += (meta_step_size * delta * x * h)
alpha = np.exp(beta)
w += delta * np.multiply(alpha, x)
h = h * np.maximum((1.0 - alpha * (x ** 2)), np.zeros((input_dim, 1))) + (alpha * delta * x)
alpha_mat[example_i, 0] = alpha[0] # relevant feature
alpha_mat[example_i, 1] = alpha[19] # irrelevant feature
x_length = np.arange(1, num_examples + 1)
plt.errorbar(x_length, alpha_mat[:, 0].flatten(), label='relevant feature')
plt.errorbar(x_length, alpha_mat[:, 1].flatten(), label='irrelevant feature')
plt.legend(loc='best')
plt.show()
plt.savefig('/content/drive/My Drive/Chicago_Project/DeltaBarDelta.png')
main(na)
Yorumlar