This document introduces machine learning concepts including supervised and unsupervised learning. It discusses preparing data for machine learning by techniques like one-hot encoding, scaling, principal component analysis (PCA), and bag-of-words representations. Code examples are provided to classify cancer data using k-nearest neighbors, cluster data with k-means, reduce dimensions with PCA, and vectorize text with bag-of-words. Finally, potential machine learning exercises are outlined like predicting user purchases, finding user clusters, and regression problems.

1. Machine Learning
Part 1. Introduction

2. Classifying images

3. Boolean
2000 LOC +
dictionary
per language
Probabilistic
20 LOC +
lots of data
all languages
machine learning

4. Types of ML

5. Supervised learning

6. Supervised learning
x
o
x
xx
x
x
o
o
o
o
o

7. Prepare
Install Anaconda: https://conda.io/docs/install/quick.html
Update: conda update condo
Create env: conda create --name <envname> python=3
Switch to env: source activate <envrname>
Install libraries: sudo pip install
numpy
scipy
matplotlib
ipython
scikit-learn
pandas
pillow

8. //load breast cancer data
from sklearn.datasets import load_breast_cancer
cancer = load_breast_cancer()
//split data into train & test
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test =
train_test_split(cancer.data,cancer.target,stratify=cancer.target,random_sta
te=66)
//use k-neighbors algorithm to perform classification
from sklearn.neighbors import KNeighborsClassifier
clf = KNeighborsClassifier(n_neighbors=3)
clf.fit(X_train,y_train

9. //predict cancer on test data
clf.predict(X_test)
//check accuracy
clf.score(X_test,y_test)

11. Unsupervised
learning

12. Unsupervised
learning
o
o
o
o
o
o
o
o
o
o
o
o
X1
X2

13. //Kmeans algorithm
from sklearn.cluster import KMeans
km = KMeans(n_clusters=2,random_state=0)
//split data into train & test
labels_km = km.fit_predict(X_train)
print(labels_km)
print(y_train)

14. Type of learning?

15. Algorithm cheat
sheet

16. Data is key
How to prepare it for ML?

17. Typical tasks
Categorical data —> one-hot-
encoding (dummy variable)
Multidimensional data —> scaling
Too many features —> Principal
Component Analysis (PCA)
Text —> bag-of-words

18. One-hot-encoding
# of flights account
# of days
since join
features
150
google,
facebook
300
gmail_parse
d_success
200 icloud 600
gmail_parse
d_success
1 live 0
3 google 1

19. One-hot-encoding
account
has_goog
le
has_faceb
ook
has_iclou
d
has_live
google,
facebook
1 1 0 0
icloud 0 0 1 0
live 0 0 0 1
google 1 0 0 –

20. //use pandas
from pandas import get_dummies
data_dummies = pd.get_dummies(data)
One-hot-encoding

21. Scaling
//minmax scaler
from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler()
scaler.fit(X_train)
//scale data
X_train_scaled = scaler.transform(X_train)
print(X_train_scaled)
print(X_train

22. PCA (eigenfaces)

23. //load Labeled Faces in the Wild dataset
from sklearn.datasets import fetch_lfw_people
people = fetch_lfw_people(min_faces_per_person=20,resize=0.7)
//display 10 faces
image_shape = people.images[0].shape
import matplotlib.pyplot as plt
fix,axes = plt.subplots(2,5, figsize=(15,8),subplot_kw={‘xticks’:(),’yticks':()})
for target,image,ax in zip(people.target,people.images,axes.ravel()):
ax.imshow(image)
ax.set_title(people.target_names[target])
plt.show()
//use plt.ion() if plot isn't displayed or create .matplotlibrc in ./.matplotlib/ with text
‘backend: TkAgg'

24. //apply k-neighbors & estimate score
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsClassifier
X_train, X_test, y_train, y_test =
train_test_split(people.data,people.target,stratify=people.target,random_state=0)
knn = KNeighborsClassifier(n_neighbors=1)
knn.fit(X_train,y_train)
knn.score(X_test,y_test)
without PCA

25. //apply PCA and then KNN
from sklearn.decomposition import PCA
pca = PCA(n_components=100,whiten=True,random_state=0).fit(X_train)
X_train_pca = pca.transform(X_train)
X_test_pca = pca.transform(X_test)
knn = KNeighborsClassifier(n_neighbors=1)
knn.fit(X_train_pca,y_train)
knn.score(X_test_pca,y_test)
with PCA

26. //display eigenfaces
fix,axes = plt.subplots(3,5,figsize=(15,12),subplot_kw={'xticks':
(),'yticks':()})
for i, (component, ax) in
enumerate(zip(pca.components_,axes.ravel())):
ax.imshow(component.reshape(image_shape),cmap='viridis')
ax.set_title("{}. component”.format((i+1)))
plt.show()
Eigenfaces

27. Eigenfaces

28. Bag-of-words

29. Bag-of-words
//vectorize
sentence = ["Hello world"]
from sklearn.feature_extraction.text import CountVectorizer
vect = CountVectorizer()
vect.fit(sentence)
//print vocabulary
vect.vocabulary_
//apply bag-of-words to sentence
bag_of_words = vect.transform(sentence)
bag_of_words.toarray()

30. Whats next?

31. Exercises
Predict user purchase (User, UserInfo,
UserSessionAction)
Find clusters of users (User, UserInfo,
UserSessionAction)
Determine if there is free wifi at the airport? (Tip)
Predicting CBP wait times at the airport
(regression)
Others?

32. Useful
CSV read: pandas.read_csv
Working with images as numpy
arrays: scikit-image
Scikit-learn.org