DST Lezione 1

Lezione 1

Books

Shamkant B. Navathe Ramez A. Elmasri. Sistemi di Basi di Dati Fondamenti e complementi. Pearson, 7^a edizione, 2018 Tom Nadeau Sam Lightstone, Toby Teorey. Physical Database Design. Morgan Kaufmann, 2007. Michael McLaughlin Scott Urman, Ron Hardman. PL/SQL programming. Oracle press, 2004.

From Navathe Elmasri book, the relevant part is Part 7

What is Data Science

What is Data Science: there is not an unique definition.

• Big Data: how big is big? why we talk about big data?
  • We have an exponential growth of the volume of data produced every day
  • The problem is that the large part of this is not structured. So, most of the data that are produced everyday are video, photo, free text. Unstructured data require a lot of memory, think of videos.
  • If data were structured, you could manage it.
  • Traditional database systems are not optimizied or suited at all for unstructured data.
  • There are some exception like BLOB: Binary Large Objects. You can store a video like a BLOB, just a large mass of bits but you have no idea what would that blob be. The bigger the file, the worst the system works. Today there is a completely different context from the one where traditional dbms were born.
• Data Science
  • Data Science can be defined as the capacity of analize a large portion data. But this is nothing more than ML or Statistics. And they overlap to a certain degree.
  • Data Science is the modern name given to experts in statistics. So, if we refer to it as DS, it is nothing new.
  • If we refer instead to databases, you already have database experties. What is now, is that you have two “legs”. One leg is analytics. So you should be able to analyzie the data. And the other leg is “storage” or “management”.
  • A data scientist must be able to do both: analyize and manage data.

Structured data vs unstructured data

In structured data it’s very simple to read data, even for a machine and information is certain. But for example, how many students are actually following the lesson? THere are two way: one is to excel table. Another way is to make a photo of students in every lesson. The information is the same, but in the second way is harder and more computational complex to interpret. There are obstruction like computers that can be mistaken by a ML Algorithm for people, and so on. Unstructured data have huge footpring, you must use the best algorithms on them and still you can’t have a certain answer. In many cases you have no alternative because you cannot collect structured data. Unstructured data is very easy to record, for example a surveillance camera. Machine Learning algorithms works but only to a certain degree. Cameras are very cheap, there are situation where you are forced to use unstructured data.

Out of the relational systems there are no “standard”. So, we can only outline general characteristics of each system and then make an example for each one. The storage systems are much simpler so it’s very easy to switch between one system and another.

Syllabus

• PART 1: Database Tuning - physical design. Depending on the data structure you use, you get major difference in query performance. Database Tuning means how to use db structure to optimizie your existing table. Database Tuning is an NP Hard problem. How to tune a database through of machine learning is one of the research fields. But right now you can’t substitute an human expert.
  • If you understand this one, the other systems are easier and it will be
  • This part must require some computation that you must do by hands.
  • One formula that evolves, do the exercise meanwhile otherwise you get lost
  • follow the course and do the exercise during the course. Don’t delay
• PART 2: No SQL Systems - easier to handle, but we need to get there.

Midterm test will be towards the end of the test. The second written test contains theoretical question and require to design a storage management for a use case.