def smoothing_eval():
#best implies least incorrect
best_one = float("inf")
best_two = 0
bayes.multinomial_factor = 0.5
smoothing_increment = 0.05
while True:
best_two = best_one
bayes.multinomial_factor += smoothing_increment
classification_rules = bayes.train_nb(train_docs, train_labels)
best_guess = bayes.classify_documents(eval_docs, classification_rules,
bayes.unique_set(train_labels))
error_cases, acc, incorrects, seperation = bayes.accuracy(
eval_labels, best_guess, eval_docs)
best_one = incorrects
log_file = open("smoothing_eval.txt", "a+")
log_file.write(
str(acc) + " " + str(seperation) + " " + str(incorrects) + " " +
str(bayes.multinomial_factor) + "\n")
log_file.close()
print(bayes.multinomial_factor)
print(str(best_one) + " " + str(best_two))
if best_one > best_two:
continue
def access():
totalAcc = 0.0
for i in range(10):
labels, samples = bayes.load_data(_filename)
trainIdx, testIdx = bayes.splitData(samples.index, 0.7)
dic = bayes.buildVocabularyList(samples)
trainVecs, trainLabels, pw = bayes.buildVecsAndLabels(
samples[trainIdx], labels[trainIdx], dic)
spamPro, pamPro = bayes.bayesian(trainVecs, trainLabels, dic)
acc = bayes.accuracy(samples[testIdx], labels[testIdx], dic, spamPro,
pamPro, pw)
totalAcc += acc
print 'accuracy = %s' % acc
averageAcc = float(totalAcc) / 10
print 'average accuracy is %f' % averageAcc
def stopword_eval(stopwords):
for i in reversed(range(0, len(stopwords))):
removed_item = stopwords[i]
stopwords.remove(removed_item)
classification_rules = bayes.train_nb(train_docs, train_labels)
best_guess = bayes.classify_documents(eval_docs, classification_rules,
bayes.unique_set(train_labels))
error_cases, acc, incorrects, seperation = bayes.accuracy(
eval_labels, best_guess, eval_docs)
log_file = open("stopword_eval.txt", "a+")
log_file.write(
str(acc) + " " + str(incorrects) + " " + " ".join(stopwords) +
"\n")
log_file.close()
def synonym_eval():
# manually set max range each time
for i in range(0, 430):
bayes.synonoyms = {}
with open("synonyms.txt", 'r') as syn_file:
raw_synonoyms = syn_file.readlines()
for index, line in enumerate(raw_synonoyms):
if index > i:
line = line.replace("\n", "").upper().split("\t")
lhs = line[1].split(",")
rhs = line[0]
for key in lhs:
bayes.synonoyms[key] = rhs
classification_rules = bayes.train_nb(train_docs, train_labels)
best_guess = bayes.classify_documents(eval_docs, classification_rules,
bayes.unique_set(train_labels))
error_cases, acc, incorrects, seperation = bayes.accuracy(
eval_labels, best_guess, eval_docs)
log_file = open("synonym_eval.txt", "a+")
log_file.write(
str(acc) + " " + str(incorrects) + " " +
" ".join(json.dumps(bayes.synonoyms)) + "\n")
log_file.close()
train_dinner_dict = bayes.fea_and_class(train_dinner)
#---calculate the mean and cov according to class------------------------
train_breakfast_summaries = bayes.normalsummarize(train_breakfast_dict)
train_lunch_summaries = bayes.normalsummarize(train_lunch_dict)
train_dinner_summaries = bayes.normalsummarize(train_dinner_dict)
#---predict the class and get a list of classes--------------------------
predictions = []
for i in range(len(val_x_list)):
if val_x_list[i][0] == "breakfast":
del val_x_list[i][0]
prediction = bayes.normalpredict(train_breakfast_dict,
train_breakfast_summaries,
val_x_list[i])
if val_x_list[i][0] == "lunch":
del val_x_list[i][0]
prediction = bayes.normalpredict(train_lunch_dict,
train_lunch_summaries,
val_x_list[i])
if val_x_list[i][0] == "dinner":
del val_x_list[i][0]
prediction = bayes.normalpredict(train_dinner_dict,
train_dinner_summaries,
val_x_list[i])
predictions.append(prediction)
#calculate the accuracy using the validation and predictions--------------
del val["Type"]
val_array = np.array(val)
val_list = val_array.tolist()
accuracy += bayes.accuracy(val_list, predictions)
average_accuracy = accuracy / 1000.0
print average_accuracy
with open("stopwords.txt", 'r') as stop_file:
bayes.filter = stop_file.readlines()
for index, line in enumerate(bayes.filter):
bayes.filter[index] = line.replace("\n", "").upper()
with open("synonyms.txt", 'r') as syn_file:
raw_synonoyms = syn_file.readlines()
for index, line in enumerate(raw_synonoyms):
line = line.replace("\n", "").upper().split("\t")
lhs = line[1].split(",")
rhs = line[0]
for key in lhs:
bayes.synonoyms[key] = rhs
print(bayes.synonoyms)
#log_file = open("stopword_eval.txt", "a+")
classification_rules = bayes.train_nb(train_docs, train_labels)
best_guess = bayes.classify_documents(eval_docs, classification_rules,
bayes.unique_set(train_labels))
error_cases, acc, incorrects, seperation = bayes.accuracy(
eval_labels, best_guess, eval_docs)
#log_file.write(str(acc) + " " + str(incorrects) + " " + " ".join(bayes.filter) + "\n")
#log_file.close()
for error in error_cases:
print(error)
pass
#print(len(error_cases))
#smoothing_eval()