We talk a lot these days about data science, and how it will pave our paths with beautiful insights and unexpected new relations and connections in our given datasets, and even across datasets.
But how to maintain the "Science" part in "Data Science"? After some time working in this field I appreciate more and more the critical thinking which has characterized the progress in science.
Hypothesis, facts, prove and/or disprove the thesis. This is how science has progressed in the past centuries. This method has been formalized by Popper and categorize as non-science all disciplines where the statements cannot be falsified. In other words, if a statement cannot be disproved, we cannot talk of science, since there is no mechanism to left to verify the solution or to prove it wrong.
When that happens the argument can still be accepted, but not scientifically accepted. Ways of accepting or refuting a non falsifiable statement are for instance based on aesthetic, authority or pragmatic or philosophical considerations. All valid but not scientific. This applies for instance to statements in the disciplines of politics, teology, ethics, etc.
Science has definitely progressed since then. For instance, Bayesian networks and statistical inductions are currently part of the arsenal of the (data) scientist weapons. But, no matter how the baseline is set, critical thinking and a rigorous method are definitely helpful in understanding the results produced by science in particular when this is based on large amount of data and computational in nature, rather than formula/model driven.
Data Science has currently many different connotations. On one side it praises the "artistry", the genius of laying out connections between disciplines and concepts. This is a truly great aspect of scientists and creativity is definitely very welcome in all data science profiles.
With the fun of creating new insights and new data golden eggs, a data scientist has to put up with those annoying criteria of reproducibility, falsifiability and peer reviewing. Sometimes these elements are postponed or left behind in name of the artistry. Granted, it's just hard to find metrics and baselines in order to compare models and data science solutions. But the scientific method has proven to be solid over the centuries and has proven to allow factual scientific discussion between scientists and a to allow selection between models based on objective agreed criteria.
7. ● (almost) everything is a number
● A few guys came with some good ideas:
Aristoteles, Galileo, Popper,
Fisher, Pearson, Bayes
What has changed in 2500 years?
14. What about it?
The shocking truth:
1) we use these concepts every day
2) we have a pre-scientific intuition of these ideas
15. Why do we bother?
New problems are related to understanding human behavior:
understand needs, desires, dreams, ambitions, cravings, and hopes.
Models have a great side effect:
they help us predicting the future.
three weapons:
Processing power: Models becomes faster: can unroll for everybody’s profiles
Sources: extract more data features, use different data.
Context: exploring information in order to understand the person.
17. How to deal with it?
Well, it’s quite simple, in a nutshell:
This is what (data) science is about:
data -> hypothesis -> validation
18. … but what we (mostly) really do is:
Use very little data
-> apply it to pre-formulated beliefs
-> come up with some “gut feeling”
Validate it:
It didn’t work? “Well, I am still right. ”
20. What’s the problem with it?
● Context
○ we could use some more data
○ insufficient feature engineering
● Add more hypotheses
○ we could explore more scenarios, “pivoting”
○ look at the problem from other angles
○ need data “artistry”
21.
22. Big data to the rescue?
Big Data is the domain which:
transforms
numbers to insights
services to experiences
23. Big data to the rescue?
by aggregating data sources
across users
across applications
across domains
24. Big data to the rescue?
in order to
providing personalized and relevant results
to the consumer of the given service
anywhere,
anytime.
25. Some small headaches
users != consumers
N=all : doesn’t mean you don’t need to clean it
Not all data is born equal
you don’t know what you don’t know
27. Some small headaches
Tough to inspect big data.
Tough to reason about big data.
representativity/bias, support, and segmentation
signal to noise ratio:
look at GFT (Google Flu Trends) for instance
28.
29. Diminishing
returns
Most of models pretty good
after a few weeks
winner added just about 5% more
after 1 year, 300 ensemble model
moral:
move on, get a new angle
30. How to compare?
You know the answer (supervised methods)
confusion matrix
ROC (Receiver Operating Characteristic)
Mean Square Error (MSE)
You don’t know the answer (unsupervised methods)
objective function
access ground truth
A/B testing
32. Beware the modeling risks
Overfitting train data
Not enough “support” in the population
Not enough features available/discovered
Not well defined objective function
34. Object functions
Many want a slice of the cake when it’s about object functions
● what the user wants
● what the community wants
● what marketing wants
● what business wants
● what finance/monetization wants
35. Data scientists
Data artists,
Data analysts
Data scientists
Data engineers
confirmatory analysis:
domain knowledge, statisticians and data analysis
exploratory analysis :
data artists/scientists
operational analysis:
data engineers , data technologists
38. What do we look in the haystack?
outliers
outliers are indicators and/or noise
groups
(Similarity metrics, PCA, SVD)
Big data as pragmatic approach to:
cheap storage
distributed computing
39. How to enjoy and compare data science?
enjoy the artistry
appreciate the genius
cross-validation
avoid falling into the trap of over-fitted models
define baseline
avoid qualitative methods
define a metric, put the models to the bench, compare results
40. Parallelism Mathematics Programming
Languages Machine Learning Statistics
Big Data Algorithms Cloud Computing
Natalino Busa
@natalinobusa
www.natalinobusa.com
Thanks !
Any questions?