Building Machine Learning Systems with Python - Richert, Coelho

More documents

Recommendations

Info

Clustering – Finding Related Posts In the previous chapter, we have learned how to find classes or categories of individual data points. With a handful of training data items that were paired with their respective classes, we learned a model that we can now use to classify future data items. We called this supervised learning, as the learning was guided by a teacher; in our case the teacher had the form of correct classifications. Let us now imagine that we do not possess those labels by which we could learn the classification model. This could be, for example, because they were too expensive to collect. What could we have done in that case? Well, of course, we would not be able to learn a classification model. Still, we could find some pattern within the data itself. This is what we will do in this chapter, where we consider the challenge of a "question and answer" website. When a user browses our site looking for some particular information, the search engine will most likely point him/her to a specific answer. To improve the user experience, we now want to show all related questions with their answers. If the presented answer is not what he/she was looking for, he/she can easily see the other available answers and hopefully stay on our site. The naive approach would be to take the post, calculate its similarity to all other posts, and display the top N most similar posts as links on the page. This will quickly become very costly. Instead, we need a method that quickly finds all related posts.
Clustering – Finding Related Posts We will achieve this goal in this chapter using clustering. This is a method of arranging items so that similar items are in one cluster and dissimilar items are in distinct ones. The tricky thing that we have to tackle first is how to turn text into something on which we can calculate similarity. With such a measurement for similarity, we will then proceed to investigate how we can leverage that to quickly arrive at a cluster that contains similar posts. Once there, we will only have to check out those documents that also belong to that cluster. To achieve this, we will introduce the marvelous Scikit library, which comes with diverse machine-learning methods that we will also use in the following chapters. Measuring the relatedness of posts From the machine learning point of view, raw text is useless. Only if we manage to transform it into meaningful numbers, can we feed it into our machine-learning algorithms such as clustering. The same is true for more mundane operations on text, such as similarity measurement. How not to do it One text similarity measure is the Levenshtein distance, which also goes by the name edit distance. Let's say we have two words, "machine" and "mchiene". The similarity between them can be expressed as the minimum set of edits that are necessary to turn one word into the other. In this case, the edit distance would be 2, as we have to add an "a" after "m" and delete the first "e". This algorithm is, however, quite costly, as it is bound by the product of the lengths of the first and second words. Looking at our posts, we could cheat by treating the whole word as characters and performing the edit distance calculation on the word level. Let's say we have two posts (let's concentrate on the title for the sake of simplicity), "How to format my hard disk" and "Hard disk format problems"; we would have an edit distance of five (removing "how", "to", "format", "my", and then adding "format" and "problems" at the end). Therefore, one could express the difference between two posts as the number of words that have to be added or deleted so that one text morphs into the other. Although we could speed up the overall approach quite a bit, the time complexity stays the same. Even if it would have been fast enough, there is another problem. The post above the word "format" accounts for an edit distance of two (deleting it first, then adding it). So our distance doesn't seem to be robust enough to take word reordering into account. [ 50 ]
Page 2 and 3:
Building Machine Learning Systems w
Page 4 and 5:
Credits Authors Willi Richert Luis
Page 6 and 7:
Luis Pedro Coelho is a Computationa
Page 8 and 9:
Maurice HT Ling completed his PhD.
Page 10 and 11:
Table of Contents Preface 1 Chapter
Page 12 and 13:
Table of Contents Tuning the instan
Page 14 and 15: Table of Contents Improving classif
Page 16 and 17: Preface You could argue that it is
Page 18 and 19: Preface What you need for this book
Page 20: Downloading the example code You ca
Page 23 and 24: Getting Started with Python Machine
Page 48 and 49: Learning How to Classify with Real-
Page 50 and 51: Chapter 2 We are using Matplotlib;
Page 52 and 53: Chapter 2 The last few lines select
Page 54 and 55: Chapter 2 error = 0.0 for ei in ran
Page 56 and 57: Chapter 2 We can play around with t
Page 58 and 59: Chapter 2 Features and feature engi
Page 60 and 61: Chapter 2 In the preceding screensh
Page 62 and 63: Chapter 2 Binary and multiclass cla
Page 66 and 67: Chapter 3 How to do it More robust
Page 68 and 69: Chapter 3 This means that the first
Page 70 and 71: Chapter 3 ... post = posts[i] ... i
Page 72 and 73: Chapter 3 If you have a clear pictu
Page 74 and 75: Chapter 3 Extending the vectorizer
Page 76 and 77: Chapter 3 0.0 >>> print(tfidf("b",
Page 78 and 79: Chapter 3 Flat clustering divides t
Page 80 and 81: Because the cluster centers are mov
Page 82 and 83: Chapter 3 'D:\\data\\379\\raw\\comp
Page 84 and 85: As we have learned previously, we w
Page 86 and 87: Chapter 3 Position Similarity Excer
Page 88: Chapter 3 But before you go there,
Page 91 and 92: Topic Modeling For those who are in
Page 93 and 94: Topic Modeling Sparsity means that
Page 95 and 96: Topic Modeling Although daunting at
Page 97 and 98: Topic Modeling … for tj,v in t:
Page 99 and 100: Topic Modeling Finally, we build th
Page 101 and 102: Topic Modeling Alternatively, we ca
Page 103 and 104: Topic Modeling Topic modeling was f
Page 105 and 106: Classification - Detecting Poor Ans
Page 115 and 116:
Classification - Detecting Poor Ans
Page 117 and 118:
Page 119 and 120:
Page 121 and 122:
Page 123 and 124:
Page 125 and 126:
Page 127 and 128:
Page 129 and 130:
Page 132 and 133:
Classification II - Sentiment Analy
Page 134 and 135:
Chapter 6 Getting to know the Bayes
Page 136 and 137:
Using Naive Bayes to classify Given
Page 138 and 139:
Chapter 6 This denotation "" leads
Page 140 and 141:
Chapter 6 Similarly, we do this for
Page 142 and 143:
Chapter 6 A quick look at the previ
Page 144 and 145:
Chapter 6 To keep our experimentati
Page 146 and 147:
Chapter 6 Y = np.zeros(Y.shape[0])
Page 148 and 149:
Chapter 6 ° ° Experiment with whe
Page 150 and 151:
Chapter 6 We have to be patient whe
Page 152 and 153:
Chapter 6 First, we define a range
Page 154 and 155:
Chapter 6 Determining the word type
Page 156 and 157:
Chapter 6 Successfully cheating usi
Page 158 and 159:
Chapter 6 Our first estimator Now w
Page 160 and 161:
Chapter 6 for d in documents: allca
Page 162 and 163:
Regression - Recommendations You ha
Page 164 and 165:
Chapter 7 The preceding graph shows
Page 166 and 167:
Chapter 7 Root mean squared error a
Page 168 and 169:
Penalized regression The important
Page 170 and 171:
Chapter 7 P greater than N scenario
Page 172 and 173:
Chapter 7 So, we can see that the d
Page 174 and 175:
Chapter 7 Fortunately, scikit-learn
Page 176 and 177:
[ 161 ] Chapter 7 The loading of th
Page 178:
Chapter 7 Summary In this chapter,
Page 181 and 182:
Regression - Recommendations Improv
Page 183 and 184:
Page 185 and 186:
Weights Regression - Recommendation
Page 187 and 188:
Page 189 and 190:
Page 191 and 192:
Page 193 and 194:
Page 196 and 197:
Classification III - Music Genre Cl
Page 198 and 199:
Chapter 9 Matplotlib provides the c
Page 200 and 201:
[ 185 ] Chapter 9
Page 202 and 203:
Chapter 9 def create_fft(fn): sampl
Page 204 and 205:
Chapter 9 ax.set_yticks(range(len(g
Page 206 and 207:
Chapter 9 On the left-hand side gra
Page 208 and 209:
[ 193 ] Chapter 9 Improving classif
Page 210 and 211:
We get the following promising resu
Page 212:
Chapter 9 Summary In this chapter,
Page 215 and 216:
Computer Vision - Pattern Recogniti
Page 217 and 218:
Page 219 and 220:
Page 221 and 222:
Page 223 and 224:
Page 225 and 226:
Page 227 and 228:
Page 229 and 230:
Page 231 and 232:
Page 233 and 234:
Page 235 and 236:
Page 237 and 238:
Dimensionality Reduction Sketching
Page 239 and 240:
Dimensionality Reduction However, t
Page 241 and 242:
Dimensionality Reduction To underst
Page 243 and 244:
Dimensionality Reduction In order t
Page 245 and 246:
Dimensionality Reduction Asking the
Page 247 and 248:
Dimensionality Reduction n_ feature
Page 249 and 250:
Dimensionality Reduction Sketching
Page 251 and 252:
Dimensionality Reduction Limitation
Page 253 and 254:
Dimensionality Reduction Now, MDS t
Page 255 and 256:
Dimensionality Reduction Of course,
Page 257 and 258:
Big(ger) Data • Your algorithms c
Page 259 and 260:
Big(ger) Data sleep(4) return 2*x @
Page 261 and 262:
Big(ger) Data Looking under the hoo
Page 263 and 264:
Big(ger) Data def write_result(ofna
Page 265 and 266:
Big(ger) Data Amazon Web Services i
Page 267 and 268:
Big(ger) Data In EC2 parlance, a ru
Page 269 and 270:
Big(ger) Data In this system, pip i
Page 271 and 272:
Big(ger) Data Keys, keys, and more
Page 273 and 274:
Big(ger) Data We can use the same j
Page 276 and 277:
Where to Learn More about Machine L
Page 278 and 279:
• Machined Learnings at http://ww
Page 280 and 281:
Index A AcceptedAnswerId attribute
Page 282 and 283:
F false negative 41 false positive
Page 284 and 285:
inary matrix of recommendations, us
Page 286:
sklearn.naive_bayes package 127 skl
Page 289 and 290:
NumPy Beginner's Guide - Second Edi
show all

Building Machine Learning Systems with Python - Richert, Coelho

Create successful ePaper yourself

Delete template?

Save as template?