def test_gradient_descent(self):
scale = Scale.scale_from_data(self.test_dataset)
# self.logreg = LogisticRegressor(self.train_dataset, self.train_labels, scale=scale)
self.logreg = LogisticRegressor(self.train_dataset, self.train_labels)
gr = GradientDescentRunner(self.logreg.get_gradient(),
len(self.samples[0]) + 1,
self.logreg.get_error_function(),
alpha=1e-8,
max_iter=300)
weights = gr.run()
self.logreg.weights = weights
mismatches = 0
predictions = [
self.logreg.predict(test_data) for test_data in self.test_dataset
]
print "All predictions: ", predictions
for test_data, label in zip(self.test_dataset, self.test_labels):
prediction = self.logreg.predict(test_data)
if prediction != label:
mismatches += 1
# print "Mismatch! Predicted ", prediction, ", True ", label
total = len(self.test_dataset)
print "total data: ", total
print "total mismatch: ", mismatches
print "percentage success: ", (
100 - float(mismatches) / float(total) * 100), "%"
def test_gradient_descent(self):
scale = Scale.scale_from_data(self.test_dataset)
# self.logreg = LogisticRegressor(self.train_dataset, self.train_labels, scale=scale)
self.logreg = LogisticRegressor(self.train_dataset, self.train_labels)
gr = GradientDescentRunner(self.logreg.get_gradient(), len(self.samples[0]) + 1,
self.logreg.get_error_function(), alpha=1e-8, max_iter=300)
weights = gr.run()
self.logreg.weights = weights
mismatches = 0
predictions = [self.logreg.predict(test_data) for test_data in self.test_dataset]
print "All predictions: ", predictions
for test_data, label in zip(self.test_dataset, self.test_labels):
prediction = self.logreg.predict(test_data)
if prediction != label:
mismatches +=1
# print "Mismatch! Predicted ", prediction, ", True ", label
total = len(self.test_dataset)
print "total data: ", total
print "total mismatch: ", mismatches
print "percentage success: ", (100 - float(mismatches) / float(total) * 100), "%"
def train_model():
training_data = pd.read_pickle("..//data//lemmatized_train_dataframe.pkl")
train_features = []
for feature in training_data[0]:
feature = feature.lower()
train_features.append(feature)
targets = []
targets.append(training_data[1])
targets.append(training_data[2])
targets.append(training_data[3])
targets.append(training_data[4])
targets.append(training_data[5])
targets.append(training_data[6])
logis = LogisticRegressor()
logis = logis.train_model(train_features, targets)
pickle.dump(logis, open("../data/logistic_model.pkl", "wb"))
def predict():
ids, comments = loadQuestionsFromTestDF()
logis = LogisticRegressor()
final_submission = open("final.csv", "w+")
for i, text in enumerate(comments):
print(ids[i])
text = str(text).lower()
text = re.sub(r'[^\w\s]', '', text)
text = remove_stop_words([text.lower()])
text = lemmatize(text[0])
results = logis.predict([text])[0]
line = str(ids[i]) + ","
for res in results:
line += str(round(res, 2)) + ","
line = line[0:len(line) - 1]
print(line)
final_submission.write(line + "\n")
class TestLogisticRegression(unittest.TestCase):
samples, labels = dataparser.parse_dataset("/Users/droy/Downloads/dataset.csv")
split_index = int(len(samples) * 0.75)
train_dataset = samples[:split_index]
train_labels = labels[:split_index]
test_dataset = samples[split_index:]
test_labels = labels[split_index:]
test_data = samples[0]
def setUp(self):
self.logreg = LogisticRegressor(self.train_dataset, self.train_labels)
def test_gradient(self):
"""Just testing that it doesn't throw any errors"""
g = self.logreg.get_gradient()
print g(ones(len(self.samples[0]) + 1))
def test_gradient_descent(self):
scale = Scale.scale_from_data(self.test_dataset)
# self.logreg = LogisticRegressor(self.train_dataset, self.train_labels, scale=scale)
self.logreg = LogisticRegressor(self.train_dataset, self.train_labels)
gr = GradientDescentRunner(self.logreg.get_gradient(), len(self.samples[0]) + 1,
self.logreg.get_error_function(), alpha=1e-8, max_iter=300)
weights = gr.run()
self.logreg.weights = weights
mismatches = 0
predictions = [self.logreg.predict(test_data) for test_data in self.test_dataset]
print "All predictions: ", predictions
for test_data, label in zip(self.test_dataset, self.test_labels):
prediction = self.logreg.predict(test_data)
if prediction != label:
mismatches +=1
# print "Mismatch! Predicted ", prediction, ", True ", label
total = len(self.test_dataset)
print "total data: ", total
print "total mismatch: ", mismatches
print "percentage success: ", (100 - float(mismatches) / float(total) * 100), "%"
def test_sckikit_performance(self):
from sklearn import linear_model
clf = linear_model.LogisticRegression()
clf.fit(self.train_dataset, self.train_labels)
predictions = clf.predict(self.test_dataset)
print "Scikit predictions: ", predictions
mismatches = 0
true_positives = 0
true_negatives = 0
false_positives = 0
false_negatives = 0
for test_data, label in zip(self.test_dataset, self.test_labels):
prediction = clf.predict(test_data)
if prediction != label:
mismatches += 1
if prediction == 1 and label == 0:
false_positives += 1
if prediction == 0 and label == 1:
false_negatives += 1
if prediction == 1 and label == 1:
true_positives += 1
if prediction == 0 and label == 0:
true_negatives += 1
# print "Mismatch! Predicted ", prediction, ", True ", label
total = len(self.test_dataset)
print "True positives: ", true_positives
print "True negatives: ", true_negatives
print "False negatives: ", false_negatives
print "False positives: ", false_positives
print "F1 metric: ", 2 * float(true_positives) \
/ float(2 * true_positives + false_positives + false_negatives)
print "MCC: ", float(true_positives * true_negatives - false_positives * false_negatives) \
/ math.sqrt((true_positives + false_positives) * (true_positives + false_negatives) \
* (true_negatives + false_positives) * (true_negatives + false_negatives))
print "total data: ", total
print "total mismatch: ", mismatches
print "percentage success: ", (100 - float(mismatches) / float(total) * 100), "%"
def test_lets_just_look_at_the_outputs(self):
gr = GradientDescentRunner(self.logreg.get_gradient(), len(self.samples[0]) + 1,
self.logreg.get_error_function(), alpha=1e-8, max_iter=300)
_, weights = gr.run_once()
self.logreg.weights = weights
predictions = [self.logreg.get_probability(d) for d in self.train_dataset]
import pprint
pprint.pprint(zip(predictions, self.train_labels))
def setUp(self):
self.logreg = LogisticRegressor(self.train_dataset, self.train_labels)
class TestLogisticRegression(unittest.TestCase):
samples, labels = dataparser.parse_dataset(
"/Users/droy/Downloads/dataset.csv")
split_index = int(len(samples) * 0.75)
train_dataset = samples[:split_index]
train_labels = labels[:split_index]
test_dataset = samples[split_index:]
test_labels = labels[split_index:]
test_data = samples[0]
def setUp(self):
self.logreg = LogisticRegressor(self.train_dataset, self.train_labels)
def test_gradient(self):
"""Just testing that it doesn't throw any errors"""
g = self.logreg.get_gradient()
print g(ones(len(self.samples[0]) + 1))
def test_gradient_descent(self):
scale = Scale.scale_from_data(self.test_dataset)
# self.logreg = LogisticRegressor(self.train_dataset, self.train_labels, scale=scale)
self.logreg = LogisticRegressor(self.train_dataset, self.train_labels)
gr = GradientDescentRunner(self.logreg.get_gradient(),
len(self.samples[0]) + 1,
self.logreg.get_error_function(),
alpha=1e-8,
max_iter=300)
weights = gr.run()
self.logreg.weights = weights
mismatches = 0
predictions = [
self.logreg.predict(test_data) for test_data in self.test_dataset
]
print "All predictions: ", predictions
for test_data, label in zip(self.test_dataset, self.test_labels):
prediction = self.logreg.predict(test_data)
if prediction != label:
mismatches += 1
# print "Mismatch! Predicted ", prediction, ", True ", label
total = len(self.test_dataset)
print "total data: ", total
print "total mismatch: ", mismatches
print "percentage success: ", (
100 - float(mismatches) / float(total) * 100), "%"
def test_sckikit_performance(self):
from sklearn import linear_model
clf = linear_model.LogisticRegression()
clf.fit(self.train_dataset, self.train_labels)
predictions = clf.predict(self.test_dataset)
print "Scikit predictions: ", predictions
mismatches = 0
true_positives = 0
true_negatives = 0
false_positives = 0
false_negatives = 0
for test_data, label in zip(self.test_dataset, self.test_labels):
prediction = clf.predict(test_data)
if prediction != label:
mismatches += 1
if prediction == 1 and label == 0:
false_positives += 1
if prediction == 0 and label == 1:
false_negatives += 1
if prediction == 1 and label == 1:
true_positives += 1
if prediction == 0 and label == 0:
true_negatives += 1
# print "Mismatch! Predicted ", prediction, ", True ", label
total = len(self.test_dataset)
print "True positives: ", true_positives
print "True negatives: ", true_negatives
print "False negatives: ", false_negatives
print "False positives: ", false_positives
print "F1 metric: ", 2 * float(true_positives) \
/ float(2 * true_positives + false_positives + false_negatives)
print "MCC: ", float(true_positives * true_negatives - false_positives * false_negatives) \
/ math.sqrt((true_positives + false_positives) * (true_positives + false_negatives) \
* (true_negatives + false_positives) * (true_negatives + false_negatives))
print "total data: ", total
print "total mismatch: ", mismatches
print "percentage success: ", (
100 - float(mismatches) / float(total) * 100), "%"
def test_lets_just_look_at_the_outputs(self):
gr = GradientDescentRunner(self.logreg.get_gradient(),
len(self.samples[0]) + 1,
self.logreg.get_error_function(),
alpha=1e-8,
max_iter=300)
_, weights = gr.run_once()
self.logreg.weights = weights
predictions = [
self.logreg.get_probability(d) for d in self.train_dataset
]
import pprint
pprint.pprint(zip(predictions, self.train_labels))
def setUp(self):
self.logreg = LogisticRegressor(self.train_dataset, self.train_labels)
import numpy as np
import random
from logistic_regression import LogisticRegressor
X = np.random.rand(380, 30)
y = [0] * 300 + [1] * 80
random.shuffle(y)
y = np.array(y)
# y = np.reshape(y,(380,1))
# print(y)
print("X:\n", X)
print("y:\n", y)
lr = LogisticRegressor(X, y)
lr.train_model(verbose=True)
# Importing libraries and modules
import pandas as pd
from utils import *
from logistic_regression import LogisticRegressor
# Reading data into variables.
X = pd.read_csv("breast_data.csv", header=None).to_numpy()
y = pd.read_csv("breast_truth.csv", header=None).to_numpy()
# Splitting data into train/test sets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)
# Creating model
model = LogisticRegressor(learning_rate=0.0001, n_iterations=100)
model.fit(X_train, y_train)
# Predict test data for evaluate model
predict_test = model.predict(X_test)
print("Accuracy of model on test data: %2.2f" % accuracy(y_test, predict_test))
