Dups NN

DupNN.py

# -*- coding: utf-8 -*-
"""
Created on Sat Jan  6 05:12:40 2018

@author: garym
"""

#import numpy as np
#import matplotlib.pyplot as plt
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import keras
from keras.models import Sequential
from keras.layers import Dense ## Makes all weights close to 0 to start 

### This is the data I have preprocessed data for ML e.g. Male = 0 / 1 rather than Gender = Male / Female
dataset = pd.read_excel(r'C:\Users\garym\Documents\PyWinAutoBlog\DD_data\DDData2.xlsx')

### set the data to predict with as all but the last column (12)
X = dataset.iloc[:, 0:12].values
### set value to predict as column 12 
y = dataset.iloc[:, 12].values

# As before use sklearn train test split 
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 0)

# This scales the features so age, for example, doesn't have too much impact as age = 70 other weights 0/1
sc = StandardScaler()
X_train = sc.fit_transform(X_train)
X_test = sc.transform(X_test)

### set up a place-holder for the neural network
classifier = Sequential()

### Create an input layer. To start with using one neuron per item used to preduct 
### relu = rectifier function. This is the best starting activation function for hidden layers
classifier.add(Dense(units = 6, kernel_initializer = 'uniform', activation = 'relu', input_dim = 12))

# Basically same as above, this is the second hidden layer
classifier.add(Dense(units = 6, kernel_initializer = 'uniform', activation = 'relu'))

# Basically same as above, this is the third hidden layer
classifier.add(Dense(units = 6, kernel_initializer = 'uniform', activation = 'relu'))

### The output layer 
### I have binary, Ledderhose yes or no so  
### Therefore I need 1 output (unit) and activation function of signmoid to give probablility
classifier.add(Dense(units = 1, kernel_initializer = 'uniform', activation = 'sigmoid'))

### Build the Neural Network 
### Use Compile function 
### adam is a type of sgd optimiser function considered pretty good for this purpose
### I use a binary loss function as binary output  
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

### Training the model
### The workhorse bit - this can take a while to run
### batch size relate to recalulating the weights, epochs is running the dataset
classifier.fit(X_train, y_train, batch_size = 10, epochs = 75)

## see how the model works when predicting the test results
y_pred = classifier.predict(X_test)
y_pred = (y_pred > 0.55)

## Create confusion matrix to see how things went
from sklearn.metrics import confusion_matrix
cm = confusion_matrix(y_test, y_pred)

## print the matrix 
print(cm)