-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
664 lines (447 loc) · 17.9 KB
/
index.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
<!DOCTYPE html>
<html lang="" xml:lang="">
<head>
<title>WWCSO AI/ML Data Challenge</title>
<meta charset="utf-8" />
<meta name="author" content="Isaac Faber" />
<meta name="date" content="2023-09-21" />
<script src="libs/header-attrs-2.24/header-attrs.js"></script>
<link rel="stylesheet" href="scrollable.css" type="text/css" />
<link rel="stylesheet" href="mtheme_max.css" type="text/css" />
<link rel="stylesheet" href="fonts_mtheme_max.css" type="text/css" />
<link rel="stylesheet" href="custom.css" type="text/css" />
</head>
<body>
<textarea id="source">
class: center, middle, inverse, title-slide
.title[
# WWCSO AI/ML Data Challenge
]
.subtitle[
## Case Study Presentation
]
.author[
### Isaac Faber
]
.date[
### 2023-09-21
]
---
class: center, middle
# Part 1 - Modeling Sales Data
---
class: center, middle
# Slides: [isaacfab.github.io](https://isaacfab.github.io)
---
# Part 1 Task Overview
.pull-left[
.font120[
* Data Cleaning and Exploring
* Holiday Effects
* Supply and Demand Effects
* Sales Prediction Modeling
* Recommended Actions
]
]
.pull-right[.center[<img src="images/agenda.png" alt="family" width="300" style />]]
---
# The Data
3 Datasets are used for this analysis. They have various amounts of complexity and preparation.
1 Sales Data
* **~400k** observations
* split between train and test (300k and ~100k obs respectively)
* weekly data ranging from `2010-02-05` to `2012-10-26`
* columns include **Store, Dept, Date, Weekly Sales, and Inventory**
* no missing or obviously corrupted data
2 Stores
* **45** observations
* columns include **Store, type, and size**
* no missing or obviously corrupted data
---
# The Data
3 Datasets are used for this analysis. They have various amounts of complexity and preparation.
3 Features
* **8190** observations
* weekly data ranging from `2010-02-05` to `2013-07-26`
* columns include **Store, Date, Temperature, Fuel Price, CPI, Unemployment, Holiday**
* Data also includes information on various promotions during holidays
* **This data is quite messy**
---
# Feature Data
.pull-left[
This dataset is quite important as it provides relevant information to weekly sales.
* **Missing Data**
* CPI and Unemployment are missing **585** observations
* Promo 1 missing **4158**
* Promo 2 missing **5269**
* Promo 3 missing **4577**
* Promo 4 missing **4726**
* Promo 5 missing **4140**
]
.pull-right[
.center[<img src="images/missing.png" alt="family" width="300" style />
]]
---
# Feature Data
With CPI and Unemployment missing data I used the suggestion to get **external data**.
After cross checking there are small discrepancies with official government figures and those provided for the case study.
For example, on just the date `2010-02-05` case study provided CPI was recorded in ranges from **126** and **214** (with a std dev of ~38) When the government reported value was **220**<sup>1</sup>
To avoid ambiguity, I have chosen to replace both missing and provided data with official government data<sup>2</sup>.
.footnote[
[1] [Consumer Price Index - All Urban Consumers](http://www.sca.isr.umich.edu/)
[2] [Unemployment from BLS](https://www.bls.gov/charts/employment-situation/civilian-unemployment-rate.htm)
]
---
# Feature Data
The case of missing data with *Promos* is tricky because the only data available is in the **future** relative to what we want to predict.
There are a number of ways to fill in (or ignore) missing data called **back casting**:
* Backward Fill
* Extrapolation
* Seasonal Decomposition
It is important to take data produced by this approach with a grain of salt.
However, in this case **seasonal decomposition** is the best approach based on the consistent nature of sales cycles.
---
# Feature Data
Back Casting Promo Values
.pull-left[.center[ Okay Model
<img src="images/tends-good.png" alt="family" width="500" style />
]]
.pull-right[.center[ Noisy Model
<img src="images/trends-bad.png" alt="family" width="500" style />
]]
* Model RMSE: **4791** is pulled up by several large outliers
* Back casting seems to reasonably capture seasonality
---
# Feature Data
In addition to CPI and Unemployment data I have added the following **external data**
* Consumer Sentiment Index
* S&P 500 market price
* VIX - volatility index
* Unemployment by age and race
---
# Sales Data
<img src="images/sales-distribution.png" alt="family" width="850" style />
---
# Sales Data
.center[<img src="images/top5-sales.png" alt="family" width="1000" style />]
---
# Sales Data
<img src="images/dept-distribution.png" alt="family" width="850" style />
---
# Sales Data
.center[<img src="images/top6-dept.png" alt="family" height="400" style/>]
---
# Holiday Effects
While the total holiday effect is clear from the previous graphs we can also model it quantitatively. The following statistical results are the product of **multiple linear regression:**
.center[<img src="images/lm-promo.png" alt="family" height="400" style/>]
The p values < .05 are of interest and indicate the present of an effect.
---
#Holdiay Effects
.center[<img src="images/lm-promo1.png" alt="family" height="400" style/>]
---
# Supply and Demand Effects
Similar to the Holiday anlsysis Supply and Demand are explored via the following statistical test using **multiple linear regression:**
.center[<img src="images/lm-sd.png" alt="family" height="400" style/>]
The p values < .05 are of interest and indicate the present of an effect.
---
# Supply and Demand Effects
.center[<img src="images/lm-inventory.png" alt="family" height="400" style/>]
---
class: center, middle
# Sales Prediction Modeling
---
# Sales Predictions
There are a few important considerations for this approach to modeling:
.pull-left[
* Using a weighted root mean squared error (RMSE) for evaluation
* A number of ML prediction algorithms will be explored
* Will be predicting **rates of change** and not actual sales
]
.pull-right[
.center[<img src="images/rate.png" alt="family" width="300" style />]
]
---
# Sales Predictions
<img src="images/sales-lag.png" alt="family" height="500" style/>
One week to the next, Sales by Store have a correlation of: **0.96**
---
# Benchmarking
Using the same technique as we did for replacing missing data a benchmark can be created;
<img src="images/season-pred.png" alt="family" height="400" style/>
With a standard Seasonal Decomposition model we get a RMSE: **5714** and a WRMSE of **5497**. For context, the Standard Deviation of Weekly Sales is ~22,000.
---
# Machine Learning Approach
To explore many different types of models the data needs to be prepared to avoid **leakage.** Some considerations are:
.pull-left[
* Only use data you 'know' when making a prediction - no time-of or future data
* Manage the lag variables to capture enough but not too much data **(T-x)**
* Based on the nature of this problem explore the following algorithms: **random forest, k-nearest neighbors, boosted trees**
* Use weighted RMSE based on holiday's for error
]
.pull-right[
.center[<img src="images/robot.png" alt="family" width="300" style />]
]
```python
def wmae(y_true, y_pred, holiday):
weights = np.where(holiday, 5, 1)
return np.sum(weights * np.abs(y_true - y_pred)) / np.sum(weights)
```
---
# Machine Learning Features
.pull-left[
Revisiting the features now that everything is consolidated
* Categorical Features: **Store, Dept, Type, day, week, month, holiday**
* Numerical Features: **return lag, inventory, size, temperature, fuel price, CPI, Consumer Sentiment, VIX, AAPL, S&P500, Unemployment (7 subsets)**
ML models often perform better when variables are **normalized**, which I did using a distribution (*Z-Value*) approach. Categorical features are **encoded** using dummy variables (one hot encoding).]
.pull-right[
.center[<img src="images/features.png" alt="family" width="300" style />]
]
---
# Machine Learning Performance
The performance of the best type of three algorithms after *feature engineering* and *hyper parameter exploration.*
.center[<img src="images/metric-table1.png" alt="family" width="800" style />]
Over 40 different ML model variations where tested.
---
# Machine Learning Performance
The performance of the best type of three algorithms after *feature engineering* and *hyper parameter exploration.*
.center[<img src="images/metric-table2.png" alt="family" width="800" style />]
Over 40 different ML model variations where tested.
---
# kNN Model Performance
<img src="images/knn.png" alt="family" width="800" style />
---
# Tree Model Performance
.pull-left[
<img src="images/gbt.png" alt="family" width="800" style />
]
.pull-right[
<img src="images/rf.png" alt="family" width="800" style />
]
---
# Feature Importance
.center[<img src="images/fi.png" alt="family" width="750" style />]
---
# Recommendations
## 1. Explore,via experimentation, the trade-off between promotions and sales on non-holidays
## 2. Holiday's provide their own demand so prioritizing inventory is best
## 3. Sales modeling should be done via a set of smaller models that aggregate - incorporate more heuristics into modeling
---
class: center, middle
# Part 2 - Research Plan: Bike+
---
# Research Plan Overview
.pull-left[
* The goal of this project is better understand demand to create **business value**
* Specifically looking at new products and how **Demand Generating (DG) events**, like launches, impact demand
* The Research Plan is above all be about **rapid experimentation and testing**
]
.pull-right[
<img src="images/researach.png" alt="family" width="500" style />
]
---
# The Research Pardigm
.center[<img src="images/decision.png" alt="family" width="900" style />]
---
# Common Tools and Team
.center[<img src="images/tools.png" alt="family" width="900" style />]
Settle quickly on a *common but flexible* technology stack that enables collaboration and rapid experimentation
---
# Data Engineering
Need a robust collection of all relevant data. Made available for exploration and production purposes.
.pull-left[
#### Internal Data
* Sales Hisotry
* Customer Demographics
* Past DG events
* Telemetry
* Relevant video/audio
#### External Data
* Market Trends
* Macro Economic Factors
* Social Media Activity
* Competitors Relevant Activity
]
.pull-right[
<img src="images/data-eng.png" alt="family" width="500" style />
]
---
# Date Science
Need a common collaborative data science environment for a team to decompose problems.
.pull-left[
#### Methodologies
* Exploratory Data Analysis
* Feature Engineering
* Model Selection
* Testing and Validation
#### Specific Focuses
* Robust statistical hypothesis testing for DG impact
* Running a set of Naive models for benchmarks
* Predicting Demand using a suite of supervised ML algorithms
* Recommend DG focus points, promotions, and inventory levels
]
.pull-right[
<img src="images/ai.png" alt="family" width="500" style />
]
---
# Date Science - Test and Evaluation
The standard for how algorithms should be evaluated will drive rapid iteration and quality.
.pull-left[
#### Methodology
* Use techniques from **Decision Analysis (DA)** for modeling choices and uncertainties (like how much inventory per store)
* ML models will be designed around the uncertainly in these DA models and prioritized based on the **Value of Data**
* The primary metric of concern will be maximizing **risk adjusted expected business value**
* Any technique to improve this metric is acceptable
]
.pull-right[
<img src="images/ai.png" alt="family" width="500" style />
]
---
# Software Engineering
.pull-left[
#### Sharing Insights
* Have a approaches to share research outcomes via UI or API
* Presentation approaches should be scaleable as needed but not at the expense of experimentation
* Should be easily tied into other organizational software engineering efforts
]
.pull-right[
<img src="images/se.png" alt="family" width="500" style />
]
---
# Conclusion & Recommendations
### 1. Create the infrastructure for rapid experimentation and deployment.
### 2. Develop and approve testing and evaluation metrics starting with risk adjusted business value (from a DA model).
### 3. Make experimentation results easily consumable by the rest of the organization.
</textarea>
<style data-target="print-only">@media screen {.remark-slide-container{display:block;}.remark-slide-scaler{box-shadow:none;}}</style>
<script src="https://remarkjs.com/downloads/remark-latest.min.js"></script>
<script>var slideshow = remark.create({
"ratio": "16:9",
"highlightStyle": "github",
"highlightLines": true,
"countIncrementalSlides": true
});
if (window.HTMLWidgets) slideshow.on('afterShowSlide', function (slide) {
window.dispatchEvent(new Event('resize'));
});
(function(d) {
var s = d.createElement("style"), r = d.querySelector(".remark-slide-scaler");
if (!r) return;
s.type = "text/css"; s.innerHTML = "@page {size: " + r.style.width + " " + r.style.height +"; }";
d.head.appendChild(s);
})(document);
(function(d) {
var el = d.getElementsByClassName("remark-slides-area");
if (!el) return;
var slide, slides = slideshow.getSlides(), els = el[0].children;
for (var i = 1; i < slides.length; i++) {
slide = slides[i];
if (slide.properties.continued === "true" || slide.properties.count === "false") {
els[i - 1].className += ' has-continuation';
}
}
var s = d.createElement("style");
s.type = "text/css"; s.innerHTML = "@media print { .has-continuation { display: none; } }";
d.head.appendChild(s);
})(document);
// delete the temporary CSS (for displaying all slides initially) when the user
// starts to view slides
(function() {
var deleted = false;
slideshow.on('beforeShowSlide', function(slide) {
if (deleted) return;
var sheets = document.styleSheets, node;
for (var i = 0; i < sheets.length; i++) {
node = sheets[i].ownerNode;
if (node.dataset["target"] !== "print-only") continue;
node.parentNode.removeChild(node);
}
deleted = true;
});
})();
// add `data-at-shortcutkeys` attribute to <body> to resolve conflicts with JAWS
// screen reader (see PR #262)
(function(d) {
let res = {};
d.querySelectorAll('.remark-help-content table tr').forEach(tr => {
const t = tr.querySelector('td:nth-child(2)').innerText;
tr.querySelectorAll('td:first-child .key').forEach(key => {
const k = key.innerText;
if (/^[a-z]$/.test(k)) res[k] = t; // must be a single letter (key)
});
});
d.body.setAttribute('data-at-shortcutkeys', JSON.stringify(res));
})(document);
(function() {
"use strict"
// Replace <script> tags in slides area to make them executable
var scripts = document.querySelectorAll(
'.remark-slides-area .remark-slide-container script'
);
if (!scripts.length) return;
for (var i = 0; i < scripts.length; i++) {
var s = document.createElement('script');
var code = document.createTextNode(scripts[i].textContent);
s.appendChild(code);
var scriptAttrs = scripts[i].attributes;
for (var j = 0; j < scriptAttrs.length; j++) {
s.setAttribute(scriptAttrs[j].name, scriptAttrs[j].value);
}
scripts[i].parentElement.replaceChild(s, scripts[i]);
}
})();
(function() {
var links = document.getElementsByTagName('a');
for (var i = 0; i < links.length; i++) {
if (/^(https?:)?\/\//.test(links[i].getAttribute('href'))) {
links[i].target = '_blank';
}
}
})();
// adds .remark-code-has-line-highlighted class to <pre> parent elements
// of code chunks containing highlighted lines with class .remark-code-line-highlighted
(function(d) {
const hlines = d.querySelectorAll('.remark-code-line-highlighted');
const preParents = [];
const findPreParent = function(line, p = 0) {
if (p > 1) return null; // traverse up no further than grandparent
const el = line.parentElement;
return el.tagName === "PRE" ? el : findPreParent(el, ++p);
};
for (let line of hlines) {
let pre = findPreParent(line);
if (pre && !preParents.includes(pre)) preParents.push(pre);
}
preParents.forEach(p => p.classList.add("remark-code-has-line-highlighted"));
})(document);</script>
<script>
slideshow._releaseMath = function(el) {
var i, text, code, codes = el.getElementsByTagName('code');
for (i = 0; i < codes.length;) {
code = codes[i];
if (code.parentNode.tagName !== 'PRE' && code.childElementCount === 0) {
text = code.textContent;
if (/^\\\((.|\s)+\\\)$/.test(text) || /^\\\[(.|\s)+\\\]$/.test(text) ||
/^\$\$(.|\s)+\$\$$/.test(text) ||
/^\\begin\{([^}]+)\}(.|\s)+\\end\{[^}]+\}$/.test(text)) {
code.outerHTML = code.innerHTML; // remove <code></code>
continue;
}
}
i++;
}
};
slideshow._releaseMath(document);
</script>
<!-- dynamically load mathjax for compatibility with self-contained -->
<script>
(function () {
var script = document.createElement('script');
script.type = 'text/javascript';
script.src = 'https://mathjax.rstudio.com/latest/MathJax.js?config=TeX-MML-AM_CHTML';
if (location.protocol !== 'file:' && /^https?:/.test(script.src))
script.src = script.src.replace(/^https?:/, '');
document.getElementsByTagName('head')[0].appendChild(script);
})();
</script>
</body>
</html>