import csv
def learn_prior(file_name, pseudo_count=0):
with open(file_name) as in_file:
training_examples = [tuple(row) for row in csv.reader(in_file)]
true_count = pseudo_count
false_count = pseudo_count
for row in training_examples[1:]:
if int(row[-1]):
true_count += 1
else:
false_count += 1
return (true_count) / (true_count + false_count)
def learn_likelihood(file_name, pseudo_count=0):
with open(file_name) as in_file:
training_examples = [tuple(row) for row in csv.reader(in_file)]
counts = []
for i in range(len(training_examples[0]) – 1):
counts.append([pseudo_count, pseudo_count])
spam_true = pseudo_count# * (len(training_examples[0]) – 1)
spam_false = pseudo_count# * (len(training_examples[0]) – 1)
for row in training_examples[1:]:
for index, value in enumerate(row[:-1]):
#print(index)
if int(value):
counts[index][int(row[-1])] += 1
#else:
# counts[index][False] += 1
if int(row[-1]):
spam_true += 1
else:
spam_false += 1
#prob_spam = learn_prior(file_name, pseudo_count)
likelihoods = []
#print(counts)
for feature_given_false, feature_given_true in counts:
#print(feature_give_false, feature_give_true)
#prob_feature = true_count / (true_count + false_count)
true_prob = feature_given_true / (spam_true + pseudo_count)#prob_feature given Spam = True
false_prob = feature_given_false / (spam_false + pseudo_count)
likelihoods.append((false_prob, true_prob))
return likelihoods
def posterior(prior, likelihood, observation):
“””Returns the posterior probability of the class variable being true, given
the observation, ie. it returns p(Class=true|observation). The argument
observation is a tuple of n booleans such that observation[i] is the
observed value (T/F) for the input feature X[i]. Prior = a real number
representing p(Class=true). likelihood = a tuple of length n where each
element is a pair of real numbers such that
likelihood[i][False] = p(X[i]=true|C=false) and
likelihood[i][True] = p(X[i]=true|C=true).”””
prior_f = 1 – prior
features_true = 1
features_false = 1
for i in range(len(observation)):
if observation[i]:
features_true *= likelihood[i][True]
features_false *= likelihood[i][False]
else:
features_true *= 1 – likelihood[i][True]
features_false *= 1 – likelihood[i][False]
features_true *= prior
features_false *= prior_f
return features_true / (features_true + features_false)
def nb_classify(prior, likelihood, input_vector):
prob_spam = posterior(prior, likelihood, input_vector)
prob_not_spam = 1 – prob_spam
if prob_spam > prob_not_spam:
return (“Spam”, prob_spam)
elif prob_spam < prob_not_spam:
return ("Not Spam", prob_not_spam)
else:
return ("Not Spam", prob_not_spam)
def accuracy(predicted_labels, correct_labels):
num_correct = 0
num_guesses = len(correct_labels)
for i in range(num_guesses):
if predicted_labels[i] == correct_labels[i]:
num_correct += 1
return num_correct / num_guesses