# 69. Machine Learning: How do you know if you have gathered enough data? By using learning rates. Do you have enough data samples to build a good machine learning model? Examining the learning rate of your model will help you decide whether it could be improved by having more data.

Here we repeat the logistic regression model on the Wisconsin Breast Cancer Diagnostic data set. We set out regularisation parameter (c) to 1, from our previous experiment, and then look at the effect of restricting our training set to varying sizes (the test set remains the same size, at 25% of our data).

We can see that as we increase our training set size the accuracy of fitting the training set reduces (it is easier to over-fit smaller data sets), and increase the accuracy of the test set. When we reach the most data we have there has not yet been a plateau in the accuracy of our test set, and the test set accuracy is significantly poorer than the training set accuracy (at out optimum regularisation for this amount of data). These two observations suggest that we would benefit from having more data for the model. If we did have more data then we should the experiment to find the optimum regularisation: generally as data set size increases, the need for regularisation reduces.

Different types of machine learning model may have different learning rates. This may influence your choice of model.

``````# import required modules

from sklearn import datasets
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

def calculate_diagnostic_performance (actual_predicted):
"""Here we truncate calulcation of results just to accuracy measurement"""

# Calculate results

test_correct = actual_predicted[:, 0] == actual_predicted[:, 1]
accuracy = np.average(test_correct)

performance = {}
performance['accuracy'] = accuracy
return performance

def chart_results(results):
x = results['n']
y1 = results['training_accuracy']
y2 = results['test_accuracy']

# Create figure
fig = plt.figure(figsize=(5,5))
ax.plot(x,y1, color='k',linestyle='solid', label = 'Training set')
ax.plot(x,y2, color='b',linestyle='dashed', label = 'Test set')
ax.set_xlabel('training set size (cases)')
ax.set_ylabel('Accuracy')
plt.title('Effect of training set size on model accuracy')
plt.legend()
plt.show()

"""Load the data set. Here we load the Breast Cancer Wisconsin (Diagnostic)
Data Set. Data could be loaded from other sources though the structure
should be compatible with thi sdata set, that is an object with the
following attribtes:
.data (holds feature data)
.feature_names (holds feature titles)
.target_names (holds outcome classification names)
.target (holds classification as zero-based number)
.DESCR (holds text-based description of data set)"""

return data_set

def normalise (X_train,X_test):
"""Normalise X data, so that training set has mean of zero and standard
deviation of one"""

# Initialise a new scaling object for normalising input data
sc=StandardScaler()
# Set up the scaler just on the training set
sc.fit(X_train)
# Apply the scaler to the training and test sets
X_train_std=sc.transform(X_train)
X_test_std=sc.transform(X_test)
return X_train_std, X_test_std

def print_diagnostic_results (performance):
"""Iterate through, and print, the performance metrics dictionary"""

print('\nMachine learning diagnostic performance measures:')
print('-------------------------------------------------')
for key, value in performance.items():
print (key,'= %0.3f' %value) # print 3 decimal places
return

def split_data (data_set, split, n):
"""Extract X and y data from data_set object, and split into tarining and
test data. Split defaults to 75% training, 25% test if not other value
passed to function"""

X=data_set.data
y=data_set.target
X_train,X_test,y_train,y_test=train_test_split(
X,y,test_size=split)
X_train = X_train[0:n]
y_train = y_train[0:n]
return X_train,X_test,y_train,y_test

def test_model(model, X, y):
"""Return predicted y given X (attributes)"""

y_pred = model.predict(X)
test_results = np.vstack((y, y_pred)).T
return test_results

def train_model (X, y):
"""Train the model """

from sklearn.linear_model import LogisticRegression
model = LogisticRegression(C=1000)
model.fit(X, y)
return model

###### Main code #######

# List of regularisation values
number_of_training_points = range(25, 450, 25)

# Set up empty lists to record results
training_accuracy = []
test_accuracy = []
n_results = []

for n in number_of_training_points:
# Repeat ml model/prediction 1000 times for each different number of runs
for i in range(1000):

# Split data into training and test sets
X_train,X_test,y_train,y_test = split_data(data_set, 0.25, n)

# Normalise data
X_train_std, X_test_std = normalise(X_train,X_test)
# Repeat test 1000x per level of c
n_results.append(n)

# Train model
model = train_model(X_train_std,y_train)

# Produce results for training set
test_results = test_model(model, X_train_std, y_train)
performance = calculate_diagnostic_performance(test_results)
training_accuracy.append(performance['accuracy'])

# Produce results for test set
test_results = test_model(model, X_test_std, y_test)
performance = calculate_diagnostic_performance(test_results)
test_accuracy.append(performance['accuracy'])

results = pd.DataFrame()
results['n'] = n_results
results['training_accuracy'] = training_accuracy
results['test_accuracy'] = test_accuracy
summary = results.groupby('n').median()
summary['n'] = list(summary.index)

print (summary)
chart_results (summary)

OUT:

training_accuracy  test_accuracy    n
n
25                 1.0       0.930070   25
50                 1.0       0.944056   50
75                 1.0       0.951049   75
100                1.0       0.951049  100
125                1.0       0.958042  125
150                1.0       0.958042  150
175                1.0       0.958042  175
200                1.0       0.958042  200
225                1.0       0.958042  225
250                1.0       0.958042  250
275                1.0       0.958042  275
300                1.0       0.958042  300
325                1.0       0.958042  325
350                1.0       0.958042  350
375                1.0       0.958042  375
400                1.0       0.951049  400
425                1.0       0.958042  425``````

# 68. Machine learning: Using regularisation to improve accuracy Many machine learning techniques include an option to fine-tune regularisation. Regularisation helps to avoid over-fitting of the model to the training set at the cost of accuracy of predication for previously unseen samples in the test set. In the logistic regression method that we have been looking at the regularisation term in the model fit is ’c’. The lower the c value the greater the regularisation. The previous code has been amended below to loop through a series of c values. For each value of c the model fit is run 100 times with different random train/test splits, and the average results are presented. Continue reading “68. Machine learning: Using regularisation to improve accuracy”

# 67. Machine learning: Adding standard diagnostic performance metrics to a ml diagnosis model

Machine learning diagnostic performance measures:
accuracy = 0.937
sensitivity = 0.933
specificity = 0.943
positive_likelihood = 16.489
negative_likelihood = 0.071
false_positive_rate = 0.057
false_negative_rate = 0.067
positive_predictive_value = 0.966
negative_predictive_value = 0.893
precision = 0.966
recall = 0.933
f1 = 0.949

# 66. Machine learning. Your first ml model! Using logistic regression to diagnose breast cancer.

Here we will use the first of our machine learning algorithms to diagnose whether someone has a benign or malignant tumour. We are using a form of logistic regression. In common to many machine learning models it incorporates a regularisation term which sacrifices a little accuracy in predicting outcomes in the training set for improved accuracy in predicting the outcomes of patients not used in the training set. Continue reading “66. Machine learning. Your first ml model! Using logistic regression to diagnose breast cancer.”

# 65. Machine learning: Feature Scaling

Many machine learning algorithms work best when numerical data for each of the features (the characteristics such as petal length and sepal length in the iris data set) are on approximately the same scale. The conversion to a similar scale is called data normalisation or data scaling. We perform this as part of out data pre-processing before training an algorithm. Continue reading “65. Machine learning: Feature Scaling”

# 63. Machine learning: Splitting data into training and test sets

To test the accuracy of a model we will test the model on data that it has not seen before. We will divide available data into two sets: a training set that the model will learn from, and a test set which will be used to test the accuracy of the model on new data. A convenient way to split the data is to use scikit-learn’s train_test_split method. This randomly divides the data between training and test sets. We may specify what proportion to keep for the test set (0.2 – 0.3 is common). Continue reading “63. Machine learning: Splitting data into training and test sets”

# 61. Machine learning: The iris data set

This is a classic ’toy’ data set used for machine learning testing is the iris data set.

The data set consists of 50 samples from each of three species of Iris (Iris setosa, Iris virginica and Iris versicolor). Four features were measured from each sample: the length and the width of the sepals and petals, in centimetres.