-
Notifications
You must be signed in to change notification settings - Fork 0
/
MLE_ND_P3_SMC.html
849 lines (848 loc) · 52.6 KB
/
MLE_ND_P3_SMC.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
<!DOCTYPE HTML>
<html>
<head>
<meta charset='utf-8'>
<meta name='viewport' content='width=device-width'>
<title>MLE_ND_P3_SMC</title>
<script src='https://sagecell.sagemath.org/static/embedded_sagecell.js'></script>
<script>$(function(){
sagecell.makeSagecell({inputLocation:'div.linked',linked:true,evalButtonText:'Run Linked Cells'});
sagecell.makeSagecell({inputLocation:'div.sage',evalButtonText:'Run'});});
</script>
<script src='https://code.highcharts.com/highcharts.js'></script>
</head>
<style>
@import 'https://fonts.googleapis.com/css?family=Orbitron|Roboto';
body {margin:5px 5px 5px 15px; background-color:ghostwhite;}
a,p {color:#5a8bbd; font-family:'Roboto';}
h1 {color:#1b2c45; font-family:'Orbitron'; text-shadow:4px 4px 4px #ccc;}
h2,h3 {color:slategray; font-family:'Orbitron'; text-shadow:4px 4px 4px #ccc;}
h4 {color:#1b2c45; font-family:'Roboto';}
.sagecell .CodeMirror-scroll {min-height:3em; max-height:60em;}
.sagecell table.table_form tr.row-a {background-color:lightgray;}
.sagecell table.table_form tr.row-b {background-color:ghostwhite;}
.sagecell table.table_form td {padding:5px 15px; color:#5a8bbd; font-family:'Roboto';}
.sagecell_sessionOutput,.sagecell_sessionOutput pre {color:#5a8bbd; font-family:'Roboto';}
</style>
<body>
<h1>🏙 Machine Learning Engineer Nanodegree
<a href='https://olgabelitskaya.github.io/README.html'>🌀 Home Page </a></h1>
<h2>Unsupervised Learning</h2>
<h1>📑 P3: Creating Customer Segments</h1>
<h2>Getting Started</h2>
<h3>Dataset</h3>
In this project, we will analyze a dataset containing data on various customers' annual spending amounts<br/>
(reported in monetary units) of diverse product categories for internal structure.<br/>
One goal of this project is to best describe the variation in the different types of customers that a wholesale distributor interacts with.<br/>
Doing so would equip the distributor with insight into how to best structure their delivery service to meet the needs of each customer.<br/>
The dataset for this project can be found on the
<a href='https://archive.ics.uci.edu/ml/datasets/Wholesale+customers'>🕸UCI Machine Learning Repository.</a><br/>
For the purposes of this project, the features <i>Channel</i> and <i>Region</i> will be excluded in the analysis — <br/>
with focus instead on the six product categories recorded for customers.
<h3>Resources</h3>
<a href='http://archive.ics.uci.edu/ml/datasets.php'>🕸UCI Machine Learning Repository </a>
<a href='https://scikit-learn.org/stable/index.html'>🕸scikit-learn. Machine Learning in Python </a>
<a href='http://seaborn.pydata.org/index.html'>🕸seaborn: statistical data visualization </a><br/>
<h3>Code Library</h3>
<div class='linked'><script type='text/x-sage'>
import warnings; warnings.filterwarnings('ignore')
from sklearn.exceptions import DataConversionWarning
warnings.filterwarnings('ignore',category=DataConversionWarning)
import numpy,pandas,pylab,seaborn,time
pylab.style.use('seaborn-pastel')
cmap=seaborn.cubehelix_palette(2,start=.1,rot=-.25,as_cmap=True)
from sklearn.model_selection import train_test_split,cross_val_score
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor
from sklearn.decomposition import PCA
from sklearn.cluster import KMeans
from sklearn.mixture import GaussianMixture
from sklearn.metrics import silhouette_score
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
# https://github.com/udacity/machine-learning/blob/master/projects/customer_segments/visuals.py
def pca_results(good_data,pca):
dimensions=['Dimension {}'.format(i)
for i in range(1,len(pca.components_)+1)]
components=pandas.DataFrame(
numpy.round(pca.components_,4),columns=good_data.keys())
components.index=dimensions
ratios=pca.explained_variance_ratio_.reshape(len(pca.components_),1)
variance_ratios=pandas.DataFrame(
numpy.round(ratios,4),columns=['Explained Variance'])
variance_ratios.index=dimensions
fig,ax=pylab.subplots(figsize=(10,5))
components.plot(ax=ax,kind='bar');
ax.set_ylabel('Feature Weights',fontsize=12)
ax.set_xticklabels(dimensions,rotation=0)
for i,ev in enumerate(pca.explained_variance_ratio_):
ax.text(i-.40,ax.get_ylim()[1]+.05,
'Explained Variance \n %.4f'%(ev))
pylab.grid(); pylab.tight_layout(); pylab.show()
return pandas.concat([variance_ratios,components],axis=int(1))
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
def cluster_results(reduced_data,preds,centers,pca_samples,alg):
predictions=pandas.DataFrame(preds,columns=['Cluster'])
plot_data=pandas.concat([predictions,reduced_data],axis=int(1))
fig,ax=pylab.subplots(figsize=(10,7))
for i,cluster in plot_data.groupby('Cluster'):
cluster.plot(
ax=ax,kind='scatter',x='Dimension 1',y='Dimension 2',
alpha=.8,edgecolors='slategray',s=30,
c=cmap((i)*1./(len(centers)-1)),label='Cluster %i'%(i));
for i,c in enumerate(centers):
ax.scatter(x=c[0],y=c[1],color='white',edgecolors='black',
alpha=1,linewidth=2,marker='o',s=250);
ax.scatter(x=c[0],y=c[1],marker='$%d$'%(i),
alpha=1,s=120,color='black');
ax.scatter(x=pca_samples[:,0],y=pca_samples[:,1],s=150,
linewidth=4,color='#ff5a8b',marker='x');
ti='%s Cluster Learning on PCA-Reduced Data\n'+\
'Centroids Marked by Numbers\n'+\
'Transformed Sample Data Marked by Red Crosses'
ax.set_title(ti%alg,fontsize=15,color='#5a8bbd')
pylab.tight_layout(); pylab.show()
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
def biplot(good_data,reduced_data,pca):
fig,ax=pylab.subplots(figsize=(10,5))
x=reduced_data.loc[:,'Dimension 1']
y=reduced_data.loc[:,'Dimension 2']
ax.scatter(x=x,y=y,facecolors='slategray',
edgecolors=cmap((x*y)**2),s=70,alpha=.4)
feature_vectors=pca.components_.T
arrow_size,text_pos=7.,5.
for i,v in enumerate(feature_vectors):
ax.arrow(0,0,arrow_size*v[0],arrow_size*v[1],
head_width=.2,head_length=.2,
linewidth=2,color='#ff5a8b')
ax.text(v[0]*text_pos,v[1]*text_pos,
good_data.columns[i],color='black',
ha='center',va='center',fontsize=12)
ax.set_xlabel('Dimension 1',fontsize=10)
ax.set_ylabel('Dimension 2',fontsize=10)
ti='PC plane with original feature projections.'
ax.set_title(ti,fontsize=15,color='#5a8bbd')
pylab.tight_layout(); pylab.show()
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
def channel_results(reduced_data,outliers,pca_samples):
path='https://raw.githubusercontent.com/OlgaBelitskaya/'+\
'machine_learning_engineer_nd009/master/'+\
'Machine_Learning_Engineer_ND_P3/'
full_data=pandas.read_csv(path+'customers.csv')
channel=pandas.DataFrame(full_data['Channel'],columns=['Channel'])
channel=channel.drop(channel.index[outliers]).reset_index(drop=True)
labeled=pandas.concat([reduced_data,channel],axis=int(1))
fig,ax=pylab.subplots(figsize=(10,5))
labels=['Hotel/Restaurant/Cafe','Retailer']
grouped=labeled.groupby('Channel')
for i,channel in grouped:
channel.plot(ax=ax,kind='scatter',
x='Dimension 1',y='Dimension 2',
color=cmap((i-1)*1./2),label=labels[i-1],s=30)
for i,sample in enumerate(pca_samples):
ax.scatter(x=sample[0],y=sample[1],s=230,linewidth=3,
color='black',marker='o',facecolors='none');
ax.scatter(x=sample[0]+.25,y=sample[1]+.3,
marker='$%d$'%(i),alpha=1,s=200,color='black')
ti='PCA-Reduced Data Labeled by `Channel`\n'+\
'Transformed Sample Data Circled'
ax.set_title(ti,fontsize=15,color='#5a8bbd')
pylab.tight_layout(); pylab.show()
</script></div><br/>
<h2>Data Exploration</h2>
In this section, we will begin exploring the data through visualizations and code to understand
how each feature is related to the others.<br/>
We will observe a statistical description of the dataset, consider the relevance of each feature, <br/>
and select a few sample data points from the dataset which you will track through the course of this project.<br/>
The dataset is composed of six important product categories:
<i>Fresh, Milk, Grocery, Frozen, Detergents_Paper</i>, and <i>Delicatessen</i>.<br/>
Consider what each category represents in terms of products we could purchase.
<h3>Data Loading</h3>
<div class='linked'><script type='text/x-sage'>
path='https://raw.githubusercontent.com/OlgaBelitskaya/'+\
'machine_learning_engineer_nd009/master/'+\
'Machine_Learning_Engineer_ND_P3/'
data=pandas.read_csv(path+'customers.csv')
data.drop(['Region','Channel'],axis=1,inplace=True)
ti='The dataset has {} samples with {} features each.'
print(ti.format(*data.shape)); data.describe()
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
data.plot.area(stacked=False,figsize=(10,5))
pylab.grid(); pylab.tight_layout(); pylab.show()
</script></div><br/>
<h3>Implementation: Selecting Samples</h3>
To get a better understanding of the customers and how their data will transform through the analysis,<br/>
it would be best to select a few sample data points and explore them in more detail.<br/>
In the code block below, add three indices of your choice to the indices list which will represent the customers to track.<br/>
It is suggested to try different sets of samples until you obtain customers that vary significantly from one another.
<div class='linked'><script type='text/x-sage'>
indices=[23,25,27]
def hex_to_rgb(chex):
chex=chex.lstrip('#'); n=len(chex)
return tuple(int(chex[i:i+n//3],16)/255. for i in range(0,n,n//3))
samples=pandas.DataFrame(data.loc[indices],
columns=data.keys()).reset_index(drop=True)
samples.index=['C0','C1','C2']; samples.loc['Mean']=samples.mean()
print ('Chosen samples of the Wholesale Customers` dataset:')
display(samples)
cols=['#1b2c45','#5a8bbd','#008b8b','#ff5a8b']
samples.T.plot(figsize=(8,4),color=cols)
pylab.grid(); pylab.tight_layout(); pylab.show()
</script></div><br/>
<h3>Question 1</h3>
<i>Consider the total purchase cost of each product category and the statistical description
of the dataset above for your sample customers.<br/>
What kind of establishment (customer) could each of the three samples you've chosen represent?</i>
<h3>Answer 1</h3>
I have built a plot for comparing the feature values with the means and a heatmap to visualize
for each customer the spending amount per feature in percentages.
<p></p>
<div class='sage'><script type='text/x-sage'>
%%html
<div id='test_highcharts' style='width:1000px; height:500px;'/><script>
Highcharts.chart('test_highcharts',{
xAxis:{categories:['Fresh','Milk','Grocery','Frozen',
'Detergents_Paper','Delicatessen']},
yAxis:{title:{text:'Value'}},
title:{style:{'color':'#5a8bbd','fontSize':'24px'},
text:'Chosen samples of the wholesale customers` dataset'},
series:[{name:'C0',color:'#1b2c45',data:[26373,36423,22019,5154,4337,16523]},
{name:'C1',color:'#5a8bbd',data:[16165,4230,7595,201,4003,57]},
{name:'C2',color:'#008b8b',data:[14276,803,3045,485,100,518]},
{name:'Mean',color:'#ff5a8b',data:[18938,13819,10886,1947,2813,5699]}] },
function(chart){
var point=chart.series.data,
text=chart.renderer.text(point.plotX+chart.plotLeft+10,
point.plotY+chart.plotTop-10).add(),
box=text.getBBox();});
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
pylab.figure(figsize=(10,5))
p_samples=samples.iloc[:].T.apply(lambda x:100.0*x/x.sum())
seaborn.heatmap(p_samples,cmap=cmap,annot=True,
annot_kws={'color':'White','size':20},fmt='.1f')
ti='Product categories in percentages for the sample customers'
pylab.title(ti,fontsize=15,color='#5a8bbd')
pylab.xticks(ha='center',fontsize=12)
pylab.yticks(fontsize=10); pylab.show()
</script></div><br/>
<b>Customer C0</b> - food-oriented supermarket:<br/>
- the wide range of products with big values; <br/>
- the significant excess of the purchasing volumes of categories <i>Fresh, Milk</i> over others;<br/>
- the purchasing volumes of categories <i>Fresh, Grocery, Milk, Delicatessen</i> are much higher than the mean values;<br/>
- the categories <i>Detergents_Paper, Frozen</i> are close to the means values.<br/>
<b>Customer C1</b> - market for the nearest neighborhood:<br/>
- the highest purchasing volumes are in the category <i>Fresh</i>;<br/>
- the purchasing volumes of all categories are close to the mean values.<br/>
<b>Customer C2</b> - vegetarian cafe or restaurant:<br/>
- the significant excess of the purchasing volumes of the category <i>Fresh</i> over others;<br/>
- the purchasing volumes of categories <i>Frozen, Grocery, Milk, Detergents_Paper</i> are lower than the mean values;<br/>
- the purchasing volumes of the category <i>Detergents_Paper</i> are extremely small.
<h3>Implementation: Feature Relevance</h3>
One interesting thought to consider is if one (or more) of the six product categories is actually relevant for understanding customer purchasing.<br/>
That is to say, is it possible to determine whether customers purchasing some amount of one category<br/>
of products will necessarily purchase some proportional amount of another category of products?<br/>
We can make this determination quite easily by training a supervised regression learner on a subset of the data with one feature removed,<br/>
and then score how well that model can predict the removed feature.<br/>
In the code block below, we will need to implement the following:<br/>
Assign <i>new_data</i> a copy of the data by removing a feature of our choice using the <i>DataFrame.drop</i> function.<br/>
Use <i>sklearn.model_selection.train_test_split</i> to split the dataset into training and testing sets.<br/>
Use the removed feature as our target label.<br/>
Set the <i>test_size</i> of 0.25 and set the <i>random_state</i> parameter.<br/>
Import the <i>Decision Tree Regressor</i>, set the <i>random_state</i> parameter, and fit the learner to the training data.<br/>
Report the prediction score of the testing set using the regressor's score function.<br/>
<h4>Experiment with <i>Delicatessen</i></h4>
<div class='linked'><script type='text/x-sage'>
new_data=data.drop('Delicatessen',axis=int(1))
target=data['Delicatessen']
new_data_target=pandas.concat([new_data,target],axis=int(1))
print ('The correlation table for the choosen feature `Delicatessen`')
pearson=new_data_target.corr(method='pearson')
corr_with_delicatessen=pearson.iloc[int(-1)][:-int(1)]
pandas.DataFrame(
corr_with_delicatessen[abs(corr_with_delicatessen)\
.argsort()[::-int(1)]]);
</script></div><br/>
<div class='linked'script type='text/x-sage'>
pylab.figure(figsize=(10,5))
seaborn.distplot(target,color='#1b2c45',bins=200,
hist_kws={'color':'SlateGrey'})
pylab.xlabel('Delicatessen',fontsize=15)
pylab.title('Customers` Annual Spending',fontsize=15,color='#5a8bbd')
pylab.grid(); pylab.tight_layout(); pylab.show()
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
def split_fit():
X_train,X_test,y_train,y_test=\
train_test_split(new_data,target,test_size=.25,random_state=1)
print ('Training and testing splitting was successful.')
regressor1=DecisionTreeRegressor(random_state=1)
regressor1.fit(X_train,y_train)
regressor2=RandomForestRegressor(random_state=1)
regressor2.fit(X_train,y_train)
score1=regressor1.score(X_test,y_test)
score2=regressor2.score(X_test,y_test)
st1='Decision Tree Regressor. '
st2='Random Forest Regressor. '
st3='The score of the prediction using the testing set is {}.'
print(st1+st3.format(score1)); print(st2+st3.format(score2))
return regressor1,regressor2
def plot_feature_importance(f,regressor1,regressor2):
D1=dict(zip(new_data,regressor1.feature_importances_))
D2=dict(zip(new_data,regressor2.feature_importances_))
pylab.figure(figsize=(10,5));
pylab.subplot(121); pylab.grid()
pylab.bar(range(len(D1)),D1.values(),alpha=.5,
color='slategrey',align='center')
pylab.xticks(range(len(D1)),D1.keys(),fontsize=7)
st='Experiment with `%s`: Feature Importances, \n'
pylab.title(st%f+'Decision Tree Regressor',fontsize=12,color='#5a8bbd')
pylab.subplot(122); pylab.grid()
pylab.bar(range(len(D2)),D2.values(),alpha=.5,
color='slategrey',align='center')
pylab.xticks(range(len(D2)),D2.keys(),fontsize=7)
pylab.title(st%f+'Random Forest Regressor',fontsize=12,color='#5a8bbd')
pylab.tight_layout(); pylab.show()
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
regressor1,regressor2=split_fit()
plot_feature_importance('Delicatessen',regressor1,regressor2)
</script></div><br/>
<h4>Experiment with <i>Grocery</i></h4>
<div class='linked'><script type='text/x-sage'>
new_data=data.drop('Grocery',axis=int(1)); target=data['Grocery']
new_data_target=pandas.concat([new_data,target],axis=int(1))
print ('The correlation table for the choosen feature `Grocery`')
pearson=new_data_target.corr(method='pearson')
corr_with_grocery=pearson.iloc[int(-1)][:-int(1)]
pandas.DataFrame(
corr_with_grocery[abs(corr_with_grocery)\
.argsort()[::-int(1)]]);
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
pylab.figure(figsize=(10,5))
seaborn.distplot(target,color='#1b2c45',bins=200,
hist_kws={'color':'SlateGrey'})
pylab.xlabel('Grocery',fontsize=15)
pylab.title('Customers` Annual Spending',fontsize=15,color='#5a8bbd')
pylab.grid(); pylab.tight_layout(); pylab.show()
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
regressor1,regressor2=split_fit()
plot_feature_importance('Grocery',regressor1,regressor2)
</script></div><br/>
<h3>Question 2</h3>
<i>Which feature did you attempt to predict? What was the reported prediction score? Is this feature is necessary for identifying customers' spending habits?</i>
<h3>Answer 2</h3>
I have chosen <i>Grocery</i>.<br/>
In this case, the reported score for predictions is much higher, and the feature has a strong correlation with at least two others (<i>Detergents_Paper</i> and <i>Milk</i>).<br/>
Therefore, it was easy to predict <i>Grocery</i> values by applying a simple regressor (<i>Decision Tree</i>).<br/>
An ensemble regressor (<i>Random Forest</i>) demonstrates better results and confirms the assumptions.<br/>
It means also that this feature is not so important for identifying spending habits of customers, and these strong<br/> relationships between variables make possible to run the procedure of dimensionality reduction.<br/>
The experiment with <i>Delicatessen</i> demonstrates the fact that it's hard to predict the values of this feature based on others, so <i>Delicatessen</i> is important for custom clustering.<br/>
<h3>Visualize Feature Distributions</h3>
To get a better understanding of the dataset, we can construct a scatter matrix of each of the six product features present in the data.<br/>
If we found that the feature we attempted to predict above is relevant for identifying a specific customer,<br/>
then the scatter matrix below may not show any correlation between that feature and the others.<br/>
Conversely, if we believe that feature is not relevant for identifying a specific customer,<br/>
the scatter matrix might show a correlation between that feature and another feature in the data.
<div class='linked'><script type='text/x-sage'>
axes=pandas.plotting.scatter_matrix(
data,alpha=.3,figsize=(10,9),diagonal='hist',c='#1b2c45',
hist_kwds={'color':'SlateGrey','bins':100})
corr=data.corr().values
for i,j in zip(*numpy.triu_indices_from(axes,k=1)):
axes[i,j].annotate('%.3f'%corr[i,j],(.8,.8),fontsize=12,
xycoords='axes fraction',ha='center',va='center')
pylab.grid(); pylab.tight_layout(); pylab.show()
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
print ('Pearson correlation coefficients')
data.corr(method='pearson')
</script></div><br/>
<h3>Question 3</h3>
<i>Are there any pairs of features which exhibit some degree of correlation?<br/>
Does this confirm or deny your suspicions about the relevance of the feature you attempted to predict?<br/>
How is the data for those features distributed?</i>
<h3>Answer 3</h3>
The highest degree of correlation has the pair <i>Grocery-Detergents_Paper</i>. These scatter plots demonstrate it very clearly.<br/>
There are some pairs else with a well-detectable correlation: <i>Milk-Grocery</i> and <i>Milk-Detergents_Paper</i>.<br/>
These facts confirm the thoughts about the relevance of the feature we attempted to predict.<br/>
We should also note that the features is not normally distributed. But the log-normal distribution looks very similar with our features.<br/>
I have created the random example of the log-normal distribution for visual comparing.
<div class='linked'><script type='text/x-sage'>
pylab.figure(figsize=(10,5)); mu,sigma=3.,1.
s=numpy.random.lognormal(mu,sigma,2000)
count,bins,ignored=pylab.hist(
s,200,align='mid',alpha=.5,color='slategrey')
x=numpy.linspace(min(bins),max(bins),2000)
pdf=(2000*numpy.exp(-(numpy.log(x)-mu)**2/(2*sigma**2))\
/(x*sigma*numpy.sqrt(2*numpy.pi)))
pylab.plot(x,pdf,linewidth=2,color='#1b2c45')
pylab.title('Example of a Random Log-Normal Distribution',
fontsize=15,color='#5a8bbd')
pylab.tight_layout(); pylab.grid(); pylab.show()
</script></div><br/>
<h2>Data Preprocessing</h2>
In this section, we will preprocessing the data to create a better representation of customers by performing a scaling on the data and detecting (and optionally removing) outliers.<br/>
Preprocessing data is often times a critical step in assuring that results you obtain from your analysis are significant and meaningful.
<h3>Implementation: Feature Scaling</h3>
If data is not normally distributed, especially if the mean and median vary significantly (indicating a large skew),
it is most often appropriate <br/>
to apply a non-linear scaling — particularly for financial data.<br/>
One way to achieve this scaling is by using a Box-Cox test, which calculates the best power transformation of the data that reduces skewness<br/>
A simpler approach which can work in most cases would be applying the natural logarithm.<br/>
Now we will do the following steps:<br/>
Assign a copy of the data to <i>log_data</i> after applying logarithmic scaling. Use the <i>numpy.log</i> function for this.<br/>
Assign a copy of the sample data to <i>log_samples</i> after applying logarithmic scaling. Again, use <i>numpy.log</i>.
<div class='linked'><script type='text/x-sage'>
log_data=numpy.log(data); log_samples=numpy.log(samples)
corr=data.corr().values
axes=pandas.plotting.scatter_matrix(
log_data,alpha=.3,figsize=(10,9),diagonal='hist',c='#1b2c45',
hist_kwds={'color':'SlateGrey','bins':100})
for i,j in zip(*numpy.triu_indices_from(axes,k=1)):
axes[i,j].annotate('%.3f'%corr[i,j],(.2,.2),fontsize=12,
xycoords='axes fraction',ha='center',va='center')
pylab.grid(); pylab.tight_layout(); pylab.show()
</script></div><br/>
<h3>Observation</h3>
After applying a natural logarithm scaling to the data, the distribution of each feature should appear much more normal.<br/>
For any pairs of features we may have identified earlier as being correlated, observe here whether that correlation is still present (and whether it is now stronger or weaker than before).
<div class='linked'><script type='text/x-sage'>
log_samples
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
print ('Pearson correlation coefficients')
log_data.corr(method='pearson')
</script></div><br/>
<h3>Implementation: Outlier Detection</h3>
Detecting outliers in the data is extremely important in the data preprocessing step of any analysis.<br/>
The presence of outliers can often skew results which take into consideration these data points.<br/>
There are many "rules of thumb" for what constitutes an outlier in a dataset.<br/>
Here, we will use the <a href='https://www.itl.nist.gov/div898/handbook/prc/section4/prc471.htm'>🕸Tukey's Method</a> for identfying outliers: an <b>outlier step</b> is calculated as 1.5 times the interquartile range (IQR).<br/>
A data point with a feature that is beyond an outlier step outside of the IQR for that feature is considered abnormal.<br/>
So the next steps are the following:<br/>
Assign the value of the 25th percentile for the given feature to Q1. Use numpy.percentile for this.<br/>
Assign the value of the 75th percentile for the given feature to Q3. Again, use numpy.percentile.<br/>
Assign the calculation of an outlier step for the given feature to step.<br/>
Optionally remove data points from the dataset by adding indices to the outliers list.
<div class='linked'><script type='text/x-sage'>
for feature in log_data.keys():
Q1,Q3=numpy.percentile(log_data[feature],25),\
numpy.percentile(log_data[feature],75)
step=(Q3-Q1)*1.5
st='Data points considered outliers for the feature `{}`:'
print(st.format(feature))
display(log_data[~((log_data[feature]>=Q1-step)&(log_data[feature]<=Q3+step))])
outliers=[65,66,75,128,154]
print('Data outliers: `{}`'.format(outliers))
good_data=log_data.drop(log_data.index[outliers]).reset_index(drop=True)
</script></div><br/>
<h3>Question 4</h3>
<i>Are there any data points considered outliers for more than one feature based on the definition above? <br/>
Should these data points be removed from the dataset?<br/>
If any data points were added to the outliers list to be removed, explain why.</i>
<h3>Answer 4</h3>
Five data points [65, 66, 75, 128, 154] can be detected as outliers for more than one feature.<br/>
It's confirmed by another method with a heatmap in the original data. I think it needs to remove them from the data.<br/>
When we apply k-means the results can be distorted by outliers: clusters are constructed with calculations of cluster-centers<br/>as the averages of all data points from this cluster, so outliers can have a great influence.<br/>
If we do not remove outliers from the dataset, they can form additional artificial clusters for the outliers which can get some data points from the real clusters.
<div class='linked'><script type='text/x-sage'>
out_percentiles=data.rank(pct=True).iloc[outliers]
pylab.figure(figsize=(10,5))
pylab.xticks(fontsize=10); pylab.yticks(fontsize=10)
seaborn.heatmap(out_percentiles,annot=True,cmap=cmap,
annot_kws={'color':'White','size':'20'})
pylab.title('Multiple Outliers Heatmap',fontsize=15,color='#5a8bbd')
pylab.tight_layout(); pylab.show()
</script></div><br/>
<h2>Feature Transformation</h2>
In this section we will use principal component analysis (PCA) to draw conclusions about the underlying structure of the wholesale customer data.<br/>
Since using PCA on a dataset calculates the dimensions which best maximize variance, we will find which compound combinations of features best describe customers.
<h3>Implementation: PCA</h3>
Now that the data has been scaled to a more normal distribution and has had any necessary outliers removed,<br/>
we can now apply PCA to the good_data to discover which dimensions about the data best maximize the variance of features involved.<br/>
In addition to finding these dimensions, PCA will also report the explained variance ratio of each dimension — how much variance within the data is explained by that dimension alone.<br/>
Note that a component (dimension) from PCA can be considered a new "feature" of the space, however it is a composition of the original features present in the data.<br/>
The next steps are the following:<br/>
Import <i>sklearn.decomposition.PCA</i> and assign the results of fitting PCA in six dimensions with <i>good_data</i> to pca.<br/>
Apply a PCA transformation of <i>log_samples</i> using <i>pca.transform</i>, and assign the results to <i>pca_samples</i>.
<div class='linked'><script type='text/x-sage'>
pca=PCA(n_components=len(good_data.keys())).fit(good_data)
pca_samples=pca.transform(log_samples)
pca_result=pca_results(good_data,pca); pca_result
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
pylab.figure(figsize=(10,4))
pylab.xlabel('Dimension',fontsize=10)
pylab.plot(numpy.arange(1,7),
numpy.cumsum(pca.explained_variance_ratio_),'-o',c='#1b2c45')
pylab.title('Explained Variance Ratio',fontsize=15,color='#5a8bbd')
pylab.grid(); pylab.tight_layout(); pylab.show()
</script></div><br/>
<h3>Question 5</h3>
<i>How much variance in the data is explained in total by the first and second principal component? What about the first four principal components?<br/>
Using the visualization provided above, discuss what the first four dimensions best represent in terms of customer spending.</i>
<h3>Answer 5</h3>
<b>Dimension 1</b>: features <i>Milk, Grocery, Detergents_Paper</i> have the highest influence, so it seems like a regular household spending on retail goods in supermarkets.<br/>
We noted in the previous section that these three features are highly correlated.<br/>
<b>Dimension 2</b>: features <i>Fresh, Frozen, Delicatessen</i> are the most important, so it can be a spending for restaurants and cafes with a wide spectrum of the menu.<br/>
<b>Dimension 3</b>: features <i>Fresh, Frozen, Delicatessen</i> are the most important also but they have different directions,<br/>so it can be a spending on retail goods in the nearest stores and markets.<br/>
<b>Dimension 4</b>: features <i>Fresh, Frozen, Delicatessen</i> are the most important again but <i>Frozen</i> has the highest influence,<br/>
so it can be a spending for restaurants and cafes with a special spectrum of the menu (fast food, for example).<br/>
The first four dimensions represent enough well in terms of customer spending because they have explained the main part of the spending variance.
<div class='linked'><script type='text/x-sage'>
print ('The first and second principal components explained {}%'\
.format(sum(pca_result['Explained Variance'][:int(2)])*100.))
print ('The first four principal components explained {}%'\
.format(sum(pca_result['Explained Variance'][:int(4)])*100.))
</script></div><br/>
<h3>PCA Observation</h3>
Let us have a look how the log-transformed sample data has changed after having a PCA transformation applied to it in six dimensions.<br/>
Observe the numerical value for the first four dimensions of the sample points.<br/>
Consider if this is consistent with our initial interpretation of the sample points.
<div class='linked'><script type='text/x-sage'>
pandas.DataFrame(
numpy.round(pca_samples,4),
columns=pca_result.index.values,
index=['C0','C1','C2','Mean'])
</script></div><br/>
<h3>Implementation: Dimensionality Reduction</h3>
When using principal component analysis, one of the main goals is to reduce the dimensionality of the data — in effect, reducing the complexity of the problem.<br/>
Dimensionality reduction comes at a cost: Fewer dimensions used implies less of the total variance in the data is being explained.<br/>
Because of this, the cumulative explained variance ratio is extremely important for knowing how many dimensions are necessary for the problem.<br/>
Additionally, if a signifiant amount of variance is explained by only two or three dimensions, the reduced data can be visualized afterwards.<br/>
The next steps are the following:<br/>
Assign the results of fitting PCA in two dimensions with <i>good_data</i> to pca.<br/>
Apply a PCA transformation of <i>good_data</i> using <i>pca.transform</i>, and assign the results to <i>reduced_data</i>.<br/>
Apply a PCA transformation of <i>log_samples</i> using <i>pca.transform</i>, and assign the results to <i>pca_samples</i>.
<div class='linked'><script type='text/x-sage'>
pca=PCA(n_components=2).fit(good_data)
pca_samples=pca.transform(log_samples)
reduced_data=pandas.DataFrame(
pca.transform(good_data),
columns=['Dimension 1','Dimension 2'])
</script></div><br/>
<h3>Dimensionality Reduction Observation</h3>
Let's see how the log-transformed sample data has changed after having a PCA transformation applied to it using only two dimensions.<br/>
Observe how the values for the first two dimensions remains unchanged when compared to a PCA transformation in six dimensions.
<div class='linked'><script type='text/x-sage'>
pandas.DataFrame(
numpy.round(pca_samples,4),
columns=['Dimension 1','Dimension 2'],
index=['C0','C1','C2','Mean'])
</script></div><br/>
<h3>Visualizing a Biplot</h3>
A biplot is a scatterplot where each data point is represented by its scores along the principal components.<br/>
The axes are the principal components (in this case <i>Dimension 1</i> and <i>Dimension 2</i>).<br/>
In addition, the biplot shows the projection of the original features along the components.<br/>
A biplot can help us interpret the reduced dimensions of the data, and discover relationships between the principal components and original features.
<div class='linked'><script type='text/x-sage'>
biplot(good_data,reduced_data,pca)
</script></div><br/>
<h3>Biplot Observation</h3>
Once we have the original feature projections (in red), it is easier to interpret the relative position of each data point in the scatterplot.<br/>
For instance, a point the lower right corner of the figure will likely correspond to a customer that spends a lot on <i>Milk</i>, <i>Grocery</i> and <i>Detergents_Paper</i>,<br/>but not so much on the other product categories.<br/>
<i>From the biplot, which of the original features are most strongly correlated with the first component?<br/>
What about those that are associated with the second component?<br/>
Do these observations agree with the pca_results plot we obtained earlier?</i><br/>
The features <i>Milk, Grocery, Detergents_Paper</i> are most strongly correlated with the first component (their projections are very close to the horizontal line).<br/>
The features <i>Fresh, Frozen, Delicatessen</i> are correlated with the second component but not so strongly (their projections are close to the vertical line).<br/>
The <i>pca_results</i> plot displays exactly the same observations.
<h2>Clustering</h2>
In this section, we will choose to use either a <b>K-Means</b> clustering algorithm or a <b>Gaussian Mixture Model</b> clustering algorithm<br/>to identify the various customer segments hidden in the data.<br/>
We will then recover specific data points from the clusters to understand their significance by transforming them back into their original dimension and scale.<br/>
<h3>Question 6</h3>
<i>What are the advantages to using a K-Means clustering algorithm?<br/>
What are the advantages to using a Gaussian Mixture Model clustering algorithm?<br/>
Given your observations about the wholesale customer data so far, which of the two algorithms will you use and why?</i>
<h3>Answer 6</h3>
The advantages of the K-Means clustering in comparing with the Gaussian Mixture Model clustering are speed and simplicity.<br/>
K-means only maintains cluster centers (linearly correlated with the feature numbers) and it will be much faster in model training.<br/>
The advantages of the Gaussian Mixture Model are a "soft" classification (indicated how likely the concrete data point belongs to the certain cluster) and<br/>
a good performance with different data distributions.<br/>
Theoretically, the Gaussian Mixture Model can do the job better thanks to its soft classification,<br/>
but the K-Means also can predict correctly without additional complexity of the model.<br/>
This project is not so large so we can try to apply both just to compare algorithms in action.
<h3>Implementation: Creating Clusters</h3>
Depending on the problem, the number of clusters that we expect to be in the data may already be known.<br/>
When the number of clusters is not known a priori, there is no guarantee that a given number of clusters best segments the data,<br/>
since it is unclear what structure exists in the data — if any.<br/>
However, we can quantify the "goodness" of a clustering by calculating each data point's silhouette coefficient.<br/>
The silhouette coefficient for a data point measures how similar it is to its assigned cluster from -1 (dissimilar) to 1 (similar).<br/>
Calculating the <b>mean silhouette coefficient</b> provides for a simple scoring method of a given clustering.<br/>
Now we need to implement the following:<br/>
Fit a clustering algorithm to the <i>reduced_data</i> and assign it to clusterer.<br/>
Predict the cluster for each data point in <i>reduced_data</i> using <i>clusterer.predict</i> and assign them to <i>preds</i>.<br/>
Find the cluster centers using the algorithm's respective attribute and assign them to <i>centers</i>.<br/>
Predict the cluster for each sample data point in <i>pca_samples</i> and assign them <i>sample_preds</i>.<br/>
Import <i>sklearn.metrics.silhouette_score</i> and calculate the <i>silhouette score</i> of <i>reduced_data</i> against <i>preds</i>.<br/>
Assign the <i>silhouette score</i> to <i>score</i> and print the result.
<h3>Question 7</h3>
<i>Report the silhouette score for several cluster numbers you tried. Of these, which number of clusters has the best silhouette score?</i>
<h3>Answer 7</h3>
I have printed scores for the K-Means clustering and the Gaussian Mixture Model.<br/>
We can see the highest scores in case of 2 clusters for both algorithms. So it should be our choice.
<div class='linked'><script type='text/x-sage'>
for n in list(range(2,12)):
clusterer=KMeans(n_clusters=n).fit(reduced_data)
preds=clusterer.predict(reduced_data)
centers=clusterer.cluster_centers_
sample_preds=clusterer.predict(pca_samples)
score=silhouette_score(reduced_data,preds)
st='For number of clusters = {}, the silhouette_score is : {}'
print(st.format(n,score))
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
for n in list(range(2,12)):
clusterer2=GaussianMixture(n_components=n,
covariance_type='full').fit(reduced_data)
preds2=clusterer2.predict(reduced_data); centers2=clusterer2.means_
sample_preds2=clusterer2.predict(pca_samples)
score2=silhouette_score(reduced_data,preds2,metric='mahalanobis')
print('For number of clusters = {}, the silhouette_score is : {}'\
.format(n,score2))
</script></div><br/>
We also can run the speed comparison for the K-Means clustering and the Gaussian Mixture Model.
<div class='linked'><script type='text/x-sage'>
n,k=100,6; kmeans_train_times=[]; gmm_train_times=[]
for k in range(2,k+1):
cum_time=0.
for i in range(n):
start=time.time()
KMeans(n_clusters=k).fit(reduced_data)
cum_time+=(time.time()-start)
kmeans_train_times.append([k,cum_time/n])
km_df=pandas.DataFrame(
kmeans_train_times,columns=['KM_Clusters','KM_Time'])
for k in range(2,k+1):
cum_time=0.
for i in range(n):
start=time.time()
GaussianMixture(n_components=k).fit(reduced_data)
cum_time+=(time.time()-start)
gmm_train_times.append([k,cum_time/n])
gmm_df=pandas.DataFrame(
gmm_train_times,columns=['GMM_Components','GMM_Time'])
times_df=km_df.join(gmm_df)
pylab.figure(figsize=(10,4))
pylab.plot(times_df.GMM_Components,times_df.GMM_Time*1000.,
'-o',c='#5a8bbd',label='GMM Train Time')
pylab.plot(times_df.GMM_Components,times_df.KM_Time*1000.,
'-o',c='#008b8b',label='Kmeans Train Time')
pylab.ylabel('Train time (in millisec.)',fontsize=10)
pylab.xlabel('Cluster/Components Used in Training',fontsize=10)
st='Training Time for Different Cluster/Component Sizes\n'+\
'Averaged Over {} Runs Per Size'
pylab.title(st.format(n),fontsize=15,color='#5a8bbd')
pylab.legend(loc='best'); pylab.grid()
pylab.tight_layout(); pylab.show()
</script></div><br/>
<h3>Cluster Visualization</h3>
Once we've chosen the optimal number of clusters for the clustering algorithm using the scoring metric above, we can now visualize the results.<br/>
Note that, for experimentation purposes, it's useful to adjust the number of clusters for the clustering algorithm to see various visualizations.<br/>
The final visualization provided should, however, correspond with the optimal number of clusters.
<div class='linked'><script type='text/x-sage'>
clusterer=KMeans(n_clusters=2).fit(reduced_data)
sample_preds=clusterer.predict(pca_samples)
preds=clusterer.predict(reduced_data)
centers=clusterer.cluster_centers_
score=silhouette_score(reduced_data,preds)
cluster_results(reduced_data,preds,centers,pca_samples,'K-Means')
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
clusterer2=GaussianMixture(
n_components=2,covariance_type='full').fit(reduced_data)
sample_preds2=clusterer2.predict(pca_samples)
preds2=clusterer2.predict(reduced_data)
centers2=clusterer2.means_; score2=silhouette_score(reduced_data,preds2)
cluster_results(reduced_data,preds2,centers2,
pca_samples,'Gaussian Mixture Model')
</script></div><br/>
<h3>Implementation: Data Recovery</h3>
Each cluster present in the visualization above has a central point.<br/>
These centers (or means) are not specifically data points from the data, but rather the averages of all the data points predicted in the respective clusters.<br/>
For the problem of creating customer segments, a cluster's center point corresponds to the average customer of that segment.<br/>
Since the data is currently reduced in dimension and scaled by a logarithm,<br/>
we can recover the representative customer spending from these data points by applying the inverse transformations.<br/>
It's time for the next steps:<br/>
Apply the inverse transform to <i>centers</i> using <i>pca.inverse_transform</i> and assign the new centers to <i>log_centers</i>.<br/>
Apply the inverse function of <i>numpy.log</i> to <i>log_centers</i> using <i>numpy.exp</i>
and assign the true centers to <i>true_centers</i>.
<div class='linked'><script type='text/x-sage'>
print ('Data Means')
pandas.DataFrame(numpy.mean(data)).T
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
log_centers=pca.inverse_transform(centers)
true_centers=numpy.exp(log_centers)
segments=['Segment {}'.format(i) for i in range(0,len(centers))]
true_centers=pandas.DataFrame(
numpy.round(true_centers),columns=data.keys(),index=segments)
print ('K-Means True Centers'); true_centers
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
true_centers=true_centers.append(data.describe().loc['mean'])
true_centers.plot(kind='bar',figsize=(10,4))
pylab.title('Comparing K-Means Centers with Data Means',
fontsize=15,color='#5a8bbd')
pylab.xticks(rotation=0,fontsize=10); pylab.grid()
pylab.tight_layout(); pylab.show()
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
log_centers2=pca.inverse_transform(centers2)
true_centers2=numpy.exp(log_centers2)
segments2=['Segment {}'.format(i) for i in range(0,len(centers2))]
true_centers2=pandas.DataFrame(
numpy.round(true_centers2),columns=data.keys(),index=segments2)
print ('Gaussian Mixture Model True Centers'); true_centers2
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
true_centers2=true_centers2.append(data.describe().loc['mean'])
true_centers2.plot(kind='bar',figsize=(10,4))
st='Comparing Gaussian Mixture Model Centers with Data Means'
pylab.title(st,fontsize=15,color='#5a8bbd')
pylab.xticks(rotation=0,fontsize=10); pylab.grid()
pylab.tight_layout(); pylab.show()
</script></div><br/>
<h3>Question 8</h3>
<i>Consider the total purchase cost of each product category for the representative data points above,<br/>
and reference the statistical description of the dataset at the beginning of this project.<br/>
What set of establishments could each of the customer segments represent?</i>
<h3>Answer 8</h3>
The values in categories are perfect indicators:<br/>
<i>Detergents_Paper, Grocery, Milk, Delicatessen</i> are higher than the mean for <b>Retail</b> and less than the mean for <b>HoReCa</b> (Hotel/Restaurant/Cafe).<br/>
<i>Fresh</i> has the highest level among others for <b>HoReCa</b> than for <b>Retail</b>.<br/>
<i>Detergents_Paper</i> has an extremely small level for <b>HoReCa</b>.
<div class='linked'><script type='text/x-sage'>
clusters=pandas.DataFrame(
2*[['','']],columns=['K-Means','Gaussian Mixture'],
index=['Segment 0','Segment 1'])
if true_centers['Fresh']['Segment 0']>true_centers['Fresh']['Segment 1']:
clusters['K-Means']['Segment 0']='HoReCa'
clusters['K-Means']['Segment 1']='Retail'
else:
clusters['K-Means']['Segment 0']='Retail'
clusters['K-Means']['Segment 1']='HoReCa'
if true_centers2['Fresh']['Segment 0']>true_centers2['Fresh']['Segment 1']:
clusters['Gaussian Mixture']['Segment 0']='HoReCa'
clusters['Gaussian Mixture']['Segment 1']='Retail'
else:
clusters['Gaussian Mixture']['Segment 0']='Retail'
clusters['Gaussian Mixture']['Segment 1']='HoReCa'
clusters
</script></div><br/>
<h3>Question 9</h3>
<i>For each sample point, which customer segment from Question 8 best represents it?<br/>
Are the predictions for each sample point consistent with this?</i>
<h3>Answer 9</h3>
<div class='linked'><script type='text/x-sage'>
clusters2=pandas.DataFrame(
3*[['','']],columns=['K-Means','Gaussian Mixture'],
index=['C0','C1','C2'])
clusters2['Cluster']=['Retail','Retail','HoReCa']
for el in ['K-Means','Gaussian Mixture']:
if clusters[el].values[0]=='Retail':
clusters2[el]=['Segment 0','Segment 0','Segment 1']
else:
clusters2[el]=['Segment 1','Segment 1','Segment 0']
clusters2
</script></div><br/>
<div class='linked'><script type='text/x-sage'>
st1='Sample point {} predicted to be in Cluster {}'
st2='The distance between sample point {} and center of Cluster {}:'
print ('K-Means')
for i,pred in enumerate(sample_preds):
print (st1.format(i,pred))
print (st2.format(i,pred))
print (samples.iloc[i]-true_centers.iloc[pred])
print ('\n\nGaussian Mixture Model')
for i,pred in enumerate(sample_preds2):
print (st1.format(i,pred))
print (st2.format(i,pred))
print (samples.iloc[i]-true_centers2.iloc[pred])
</script></div><br/>
<h2>Conclusion</h2>
In this final section, we will investigate ways that we can make use of the clustered data.<br/>
First, we will consider how the different groups of customers, the customer segments, may be affected differently by a specific delivery scheme.<br/>
Next, we will consider how giving a label to each customer (which segment that customer belongs to) can provide for additional features about the customer data.<br/>
Finally, we will compare the customer segments to a hidden variable present in the data, to see whether the clustering identified certain relationships.
<h3>Question 10</h3>
<i>Companies will often run A/B tests when making small changes to their products or services to determine whether making that change will affect its customers positively or negatively.<br/>
The wholesale distributor is considering changing its delivery service from currently 5 days a week to 3 days a week.<br/>
However, the distributor will only make this change in delivery service for customers that react positively.<br/>
How can the wholesale distributor use the customer segments to determine which customers, if any, would react positively to the change in delivery service?</i>
<h3>Answer 10</h3>
If we reduce the delivery frequency for Retail it's possible to have a positive reaction:<br/>
they do not order a lot of fresh food, it can reduce the transport cost, and their spending values are more predictable and regular.<br/>
But some reactions could be negative: many supermarkets try to minimize inventory to save cost.<br/>
The customers in HoReCa (Hotel/Restaurant/Cafe) can react positively if they have enough places for saving food.<br/>
But a negative reaction also can be: some of them do not have enough space for saving, but they need fresh food for their business.<br/>
For both segments, a negative effect is possible.<br/>
Only an A/B test can detect if actually true.<br/>
We can run the A/B test with randomly selected samples for each cluster.<br/>
It is possible to have absolutely different tendencies in the test results.<br/>
One segment could have the significant effect, another segment could not demonstrate it.<br/>
Their spending habits are not similar. If we mixed them we could lose information about the real effect.
<h3>Question 11</h3>
<i>Additional structure is derived from originally unlabeled data when using clustering techniques.<br/>
Since each customer has a customer segment it best identifies with (depending on the clustering algorithm applied),<br/>
we can consider 'customer segment' as an engineered feature for the data.<br/>
Assume the wholesale distributor recently acquired ten new customers and each provided estimates for anticipated annual spending of each product category.<br/>
Knowing these estimates, the wholesale distributor wants to classify each new customer to a customer segment to determine the most appropriate delivery service.<br/>
How can the wholesale distributor label the new customers using only their estimated product spending and the customer segment data?</i>
<h3>Answer 11</h3>
There are several ways for using and improving analytic predictions for labels:<br/>
- the clustering algorithms K-Means and GMMs updated for the new data points to include them into the certain clusters;<br/>
- supervised learning algorithms to detect differences between clusters, cluster labels can be just a target in the model predictions for new data points.<br/>
<h3>Visualizing Underlying Distributions</h3>
At the beginning of this project, it was discussed that the <i>Channel</i> and 'Region' features would be excluded from the dataset<br/>so that the customer product categories were emphasized in the analysis.<br/>
By reintroducing the <i>Channel</i> feature to the dataset, an interesting structure emerges when considering<br/>the same PCA dimensionality reduction applied earlier to the original dataset.<br/>
Let's see how each data point is labeled either <i>HoReCa</i> (Hotel/Restaurant/Cafe) or <i>Retail</i> the reduced space.<br/>
In addition, we will find the sample points are circled in the plot, which will identify their labeling.
<div class='linked'><script type='text/x-sage'>
channel_results(reduced_data,outliers,pca_samples[:3])
</script></div><br/>
<h3>Question 12</h3>
<i>How well does the clustering algorithm and number of clusters you've chosen compare to this underlying distribution of Hotel/Restaurant/Cafe customers to Retailer customers?<br/>
Are there customer segments that would be classified as purely <i>Retailers</i> or <i>Hotels/Restaurants/Cafes</i> by this distribution?<br/>
Would you consider these classifications as consistent with your previous definition of the customer segments?</i>
<h3>Answer 12</h3>
The number of clusters is consistent with the underlying distribution so both clustering algorithms work enough well.<br/>
As we can see the K-Means simplicity does not affect the efficiency in this case.<br/>
The customer segments classified as purely <i>Retailers</i> or <i>Hotels/Restaurants/Cafes</i> on the left side and on the right side accordingly.<br/>
The algorithmic classification in the majority of cases is in line with the real results.<br/>
We should note that for both algorithms borders of clusters are more clearly detectable.<br/>
Differences between actual and algorithmic classification may indicate a lack of the number of clusters for this market.<br/> Perhaps it needs a greater number to detect spending habits.
<div class='linked'><script type='text/x-sage'>
channel_data=pandas.read_csv(path+'customers.csv')
channel_data=channel_data.drop(
channel_data.index[outliers]).reset_index(drop=True)
if clusters2['K-Means'].values[0]=='Segment 0': v11,v12=0,1
else: v11,v12=1,0
if clusters2['Gaussian Mixture'].values[0]=='Segment 0': v21,v22=0,1
else: v21,v22=1,0
df=numpy.where(channel_data['Channel']==2,v11,v12)
percentage_kmeans=sum(df==preds)/float(len(preds))
df2=numpy.where(channel_data['Channel']==2,v21,v22)
percentage_gaussian_mixture=sum(df2==preds2)/float(len(preds2))
print ('K-Means')
print ('Percentage of correctly classified customers: {:.2%}'\
.format(percentage_kmeans))
print ('\nGaussian Mixture Model')
print ('Percentage of correctly classified customers: {:.2%}'\
.format(percentage_gaussian_mixture))
</script></div><br/>
<h3>Reflections</h3>
It would be interesting to repeat the steps of this project in R and compare the results.<br/>
Or try animation for clustering algorithms to compare the borders of clusters with real data labels during the learning process.
<h3>Additional Code Cell</h3>
<div class='linked'><script type='text/x-sage'>
</script></div><br/>
</body>
</html>