Monday, January 16, 2006


1. A good reference is the Tom Mitchell: Machine learning, A gentle tutorial on EM algorithm.

2. What is a mixture model?
Stolen from Wikipedia

A mixture model is a model in which the independent variables are measured as fractions of a total. For example, suppose researchers are trying to find the optimal mixture of ingredients for a fruit punch consisting of grape juice, mango juice, and pineapple juice. A mixture model is suitable here because the results of the taste tests will not depend on the amount of ingredients used to make the batch but rather on the fraction of each ingredient present in the punch. The sum of the mixture components is always 100%, and a mixture model takes this restriction into account.

Another definition
A mixture model can also be a formalism for modeling a probability density function as a sum of parameterized functions. In mathematical terms,

where pX(x) is the modeled probability distribution function, K is the number of components in the mixture model, and ak is mixture proportion of component k. By definition, 0 < ak < 1 for all and .

h(x | λk) is a probability distribution parameterized by λk.
Mixture models are often used when we know h(x) and we can sample from pX(x), but we would like to determine the ak and λk values. Such situations can arise in studies in which we sample from a population that is composed of several distinct subpopulations.

Two common approaches for estimation in mixture models
It's common to think of mixture modeling (under the second definition) as a missing data problem. One way to understand this is to assume that the data points under consideration have "membership" in one of the distributions we are using to model the data. When we start, this membership is unknown, or missing. The job of estimation is to devise appropriate parameters for the model functions we choose, with the connection to the data points being represented as their membership in the individual model distributions.

Expectation maximization
The Expectation-maximization algorithm is one way to compute the missing memberships of data points in our chosen distribution model. It is an iterative procedure, where we start with initial parameters for our model distribution (the ak's and λk's of the model listed above). The estimation process proceeds iteratively in two steps, the Expectation Step, and the Maximization Step.

The expectation step
With initial guesses for the parameters in our mixture model, we compute "partial membership" of each data point in each constituent distribution. This is done by calculating expectation values for the membership variables of each data point. An example will provide some clarity. Let's consider a simple example. We have a collection of data points xi that can be modeled as coming from a sum of two Gaussian distributions. The probability expression for our model is

Where f is the mixing coefficient in [0,1] (ak = 2 = f and ak = 1 = 1 − f), and we assume σ is known and constant. For each of our data points xi, we can compute a membership value for each of the two Gaussians as follows

and similarly for y2,i.

In the case of a Gaussian mixture model,

The maximization step
With our expectation values in hand for group membership, we can recompute plug-in estimates of our distribution parameters. For the mixing coefficient f this is simply the fractional membership of all data points in the second Gaussian.

where N is the total number of data points. For μ1,

With new estimates for f and the μ's, we proceed back to the Expectation step to recompute new membership values. The procedure is repeated until there is no further change in the mixture model parameters.

Saturday, January 07, 2006

Java Operators

* << left-shift, >> right-shift, >>> unsigned right-shift
* only used on integer values
* binary numeric promotion is not performed on the operands; instead unary promotion is performed on each operand separately (JLS §15.19)
* both operands are individually promoted to int if their type is byte, short or char
* a long shift operator does not force a left-hand int value promotion to long (JLS§5.6.1)
* left-associative
* left-hand operator represents the number to be shifted
* right-hand operator specifies the shift distance

value << 2 // 2 is the distance to be shifted

* when the value to be shifted (left-operand) is an int, only the last 5 digits of the right-hand operand are used to perform the shift. The actual size of the shift is the value of the right-hand operand masked by 31 (0x1f). ie the shift distance is always between 0 and 31 (if shift value is > 32 shift is 32%value)

35 00000000 00000000 00000000 00100011
31 -> 0x1f 00000000 00000000 00000000 00011111
& -----------------------------------
Shift value 00000000 00000000 00000000 00000011 -> 3

-29 11111111 11111111 11111111 11100011
31 -> 0x1f 00000000 00000000 00000000 00011111
& -----------------------------------
Shift value 00000000 00000000 00000000 00000011 -> 3

* when the value to be shifted (left-operand) is a long, only the last 6 digits of the right-hand operand are used to perform the shift. The actual size of the shift is the value of the right-hand operand masked by 63 (0x3D) ie the shift distance is always between 0 and 63 (if shift value is greater than 64 shift is 64%value)
* the shift occurs at runtime on a bit-by-bit basis

Left-shift << (JLS §15.19)

* bits are shifted to the left based on the value of the right-operand
* new right hand bits are zero filled
* equivalent to left-operand times two to the power of the right-operand
For example, 16 << 5 = 16 * 25 = 512

Decimal 16 00000000000000000000000000010000

Left-shift 5 00000000000000000000000000010000
fill right 0000000000000000000000000001000000000
discard left 00000000000000000000001000000000

* the sign-bit is shifted to the left as well, so it can be dropped off or a different sign can replace it

Right-shift >> (JLS §15.19)

* bits are shifted to the right based on value of right-operand
* new left hand bits are filled with the value of the left-operand high-order bit therefore the sign of the left-hand operator is always retained
* for non-negative integers, a right-shift is equivalent to dividing the left-hand operator by two to the power of the right-hand operator
For example: 16 >> 2 = 16 / 22 = 4

Decimal 16 00000000000000000000000000010000

Right-shift 2 00000000000000000000000000010000
fill left 00000000000000000000000000000100
discard right 00000000000000000000000000000100 -> Decimal 4

Decimal -16 11111111111111111111111111110000

Right-shift 2 11111111111111111111111111110000
fill left 1111111111111111111111111111110000
discard right 11111111111111111111111111111100 -> Decimal -4

Unsigned right-shift >>> (JLS §15.19)

* identical to the right-shift operator only the left-bits are zero filled
* because the left-operand high-order bit is not retained, the sign value can change
* if the left-hand operand is positive, the result is the same as a right-shift
* if the left-hand operand is negative, the result is equivalent to the left-hand operand right-shifted by the number indicated by the right-hand operand plus two left-shifted by the inverted value of the right-hand operand
For example: -16 >>> 2 = (-16 >> 2 ) + ( 2 << ~2 ) = 1,073,741,820

Decimal 16 00000000000000000000000000010000

Right-shift 2 00000000000000000000000000010000
fill left 00000000000000000000000000000100
discard right 00000000000000000000000000000100 -> Decimal 4

Decimal -16 11111111111111111111111111110000

>>> 2 11111111111111111111111111110000
fill left 0011111111111111111111111111110000
discard right 00111111111111111111111111111100

Gaussian Mixture Models

Gaussian Mixture Models are universal approximators of densities.
Gaussian Mixture Models are trained by maximum likelihood using Expectation Maximization algorithm

Friday, January 06, 2006

Oracle DBA Interview Question

from Koopmann)

1. Explain the difference between a hot backup and a cold backup and the benefits associated with each.

A hot backup is basically taking a backup of the database while it is still up and running and it must be in archive log mode. A cold backup is taking a backup of the database while it is shut down and does not require being in archive log mode. The benefit of taking a hot backup is that the database is still available for use while the backup is occurring and you can recover the database to any point in time. The benefit of taking a cold backup is that it is typically easier to administer the backup and recovery process. In addition, since you are taking cold backups the database does not require being in archive log mode and thus there will be a slight performance gain as the database is not cutting archive logs to disk.

2. You have just had to restore from backup and do not have any control files. How would you go about bringing up this database?

I would create a text based backup control file, stipulating where on disk all the data files where and then issue the recover command with the using backup control file clause.

3. How do you switch from an init.ora file to a spfile?

Issue the create spfile from pfile command.

4. Explain the difference between a data block, an extent and a segment.

A data block is the smallest unit of logical storage for a database object. As objects grow they take chunks of additional storage that are composed of contiguous data blocks. These groupings of contiguous data blocks are called extents. All the extents that an object takes when grouped together are considered the segment of the database object.

5. Give two examples of how you might determine the structure of the table DEPT.

Use the describe command or use the dbms_metadata.get_ddl package.

6. Where would you look for errors from the database engine?

In the alert log.

7. Compare and contrast TRUNCATE and DELETE for a table.

Both the truncate and delete command have the desired outcome of getting rid of all the rows in a table. The difference between the two is that the truncate command is a DDL operation and just moves the high water mark and produces a now rollback. The delete command, on the other hand, is a DML operation, which will produce a rollback and thus take longer to complete.

8. Give the reasoning behind using an index.

Faster access to data blocks in a table.

9. Give the two types of tables involved in producing a star schema and the type of data they hold.

Fact tables and dimension tables. A fact table contains measurements while dimension tables will contain data that will help describe the fact tables.

10. . What type of index should you use on a fact table?

A Bitmap index.

11. Give two examples of referential integrity constraints.

A primary key and a foreign key.

12. A table is classified as a parent table and you want to drop and re-create it. How would you do this without affecting the children tables?

Disable the foreign key constraint to the parent, drop the table, re-create the table, enable the foreign key constraint.

13. Explain the difference between ARCHIVELOG mode and NOARCHIVELOG mode and the benefits and disadvantages to each.

ARCHIVELOG mode is a mode that you can put the database in for creating a backup of all transactions that have occurred in the database so that you can recover to any point in time. NOARCHIVELOG mode is basically the absence of ARCHIVELOG mode and has the disadvantage of not being able to recover to any point in time. NOARCHIVELOG mode does have the advantage of not having to write transactions to an archive log and thus increases the performance of the database slightly.

14. What command would you use to create a backup control file?

Alter database backup control file to trace.

15. Give the stages of instance startup to a usable state where normal users may access it.

STARTUP NOMOUNT - Instance startup

STARTUP MOUNT - The database is mounted

STARTUP OPEN - The database is opened

16. What column differentiates the V$ views to the GV$ views and how?

The INST_ID column which indicates the instance in a RAC environment the information came from.

17. How would you go about generating an EXPLAIN plan?

Create a plan table with utlxplan.sql.

Use the explain plan set statement_id = 'tst1' into plan_table for a SQL statement

Look at the explain plan with utlxplp.sql or utlxpls.sql

18. How would you go about increasing the buffer cache hit ratio?

Use the buffer cache advisory over a given workload and then query the v$db_cache_advice table. If a change was necessary then I would use the alter system set db_cache_size command.

19. Explain an ORA-01555

You get this error when you get a snapshot too old within rollback. It can usually be solved by increasing the undo retention or increasing the size of rollbacks. You should also look at the logic involved in the application getting the error message.

20. Explain the difference between $ORACLE_HOME and $ORACLE_BASE.

ORACLE_BASE is the root directory for oracle. ORACLE_HOME located beneath ORACLE_BASE is where the oracle products reside.

21. How would you determine the time zone under which a database was operating?

select DBTIMEZONE from dual;

22. Explain the use of setting GLOBAL_NAMES equal to TRUE.

Setting GLOBAL_NAMES dictates how you might connect to a database. This variable is either TRUE or FALSE and if it is set to TRUE it enforces database links to have the same name as the remote database to which they are linking.

23. What command would you use to encrypt a PL/SQL application?


24. Explain the difference between a FUNCTION, PROCEDURE and PACKAGE.

A function and procedure are the same in that they are intended to be a collection of PL/SQL code that carries a single task. While a procedure does not have to return any values to the calling application, a function will return a single value. A package on the other hand is a collection of functions and procedures that are grouped together based on their commonality to a business function or application.

25. Explain the use of table functions.

Table functions are designed to return a set of rows through PL/SQL logic but are intended to be used as a normal table or view in a SQL statement. They are also used to pipeline information in an ETL process.

26. Name three advisory statistics you can collect.

Buffer Cache Advice, Segment Level Statistics, & Timed Statistics

27. Where in the Oracle directory tree structure are audit traces placed?

In unix $ORACLE_HOME/rdbms/audit, in Windows the event viewer

28. Explain materialized views and how they are used.

Materialized views are objects that are reduced sets of information that have been summarized, grouped, or aggregated from base tables. They are typically used in data warehouse or decision support systems.

29. When a user process fails, what background process cleans up after it?


30. What background process refreshes materialized views?

The Job Queue Processes.

31. How would you determine what sessions are connected and what resources they are waiting for?


32. Describe what redo logs are.

Redo logs are logical and physical structures that are designed to hold all the changes made to a database and are intended to aid in the recovery of a database.

33. How would you force a log switch?


34. Give two methods you could use to determine what DDL changes have been made.

You could use Logminer or Streams

35. What does coalescing a tablespace do?

Coalescing is only valid for dictionary-managed tablespaces and de-fragments space by combining neighboring free extents into large single extents.

36. What is the difference between a TEMPORARY tablespace and a PERMANENT tablespace?

A temporary tablespace is used for temporary objects such as sort structures while permanent tablespaces are used to store those objects meant to be used as the true objects of the database.

37. Name a tablespace automatically created when you create a database.

The SYSTEM tablespace.

38. When creating a user, what permissions must you grant to allow them to connect to the database?

Grant the CONNECT to the user.

39. How do you add a data file to a tablespace?


40. How do you resize a data file?


41. What view would you use to look at the size of a data file?


42. What view would you use to determine free space in a tablespace?


43. How would you determine who has added a row to a table?

Turn on fine grain auditing for the table.

44. How can you rebuild an index?


45. Explain what partitioning is and what its benefit is.

Partitioning is a method of taking large tables and indexes and splitting them into smaller, more manageable pieces.

46. You have just compiled a PL/SQL package but got errors, how would you view the errors?


47. How can you gather statistics on a table?

The ANALYZE command.

48. How can you enable a trace for a session?



49. What is the difference between the SQL*Loader and IMPORT utilities?

These two Oracle utilities are used for loading data into the database. The difference is that the import utility relies on the data being produced by another Oracle utility EXPORT while the SQL*Loader utility allows data to be loaded that has been produced by other utilities from different data sources just so long as it conforms to ASCII formatted or delimited files.

50. Name two files used for network connection to a database.

Technical - UNIX

Every DBA should know something about the operating system that the database will be running on. The questions here are related to UNIX but you should equally be able to answer questions related to common Windows environments.

1. How do you list the files in an UNIX directory while also showing hidden files?

ls -ltra

2. How do you execute a UNIX command in the background?

Use the "&"

3. What UNIX command will control the default file permissions when files are created?


4. Explain the read, write, and execute permissions on a UNIX directory.

Read allows you to see and list the directory contents.

Write allows you to create, edit and delete files and subdirectories in the directory.

Execute gives you the previous read/write permissions plus allows you to change into the directory and execute programs or shells from the directory.

5. the difference between a soft link and a hard link?

A symbolic (soft) linked file and the targeted file can be located on the same or different file system while for a hard link they must be located on the same file system.

6. Give the command to display space usage on the UNIX file system.

df -lk

7. Explain iostat, vmstat and netstat.

Iostat reports on terminal, disk and tape I/O activity.

Vmstat reports on virtual memory statistics for processes, disk, tape and CPU activity.

Netstat reports on the contents of network data structures.

8. How would you change all occurrences of a value using VI?

Use :%s///g

9. Give two UNIX kernel parameters that effect an Oracle install


10. Briefly, how do you install Oracle software on UNIX.

Basically, set up disks, kernel parameters, and run orainst.


Why was i always afraid of PL/SQL?
Simple example from Oracles site
PROCEDURE debit_account (acct_id INTEGER, debit_amount REAL) IS
old_balance REAL;
new_balance REAL;
overdrawn EXCEPTION;
SELECT bal INTO old_balance FROM accts
WHERE acct_no = acct_id;
new_balance := old_balance - debit_amount;
IF new_balance < 0 THEN
RAISE overdrawn;
UPDATE accts SET bal = new_balance
WHERE acct_no = acct_id;
WHEN overdrawn THEN
-- handle the error
END debit_account;

Maximum Likelihood

Stolen from wikipedia :)

What is likelihood? ( from mathworld)
Likelihood is the hypothetical probability that an event that has already occurred would yield a specific outcome. The concept differs from that of a probability in that a probability refers to the occurrence of future events, while a likelihood refers to past events with known outcomes.

Maximum likelihood estimation (MLE) is a popular statistical method used to make inferences about parameters of the underlying probability distribution of a given data set.

What is probability distribution?
In mathematics, a probability distribution assigns to every interval of the real numbers a probability, so that the probability axioms are satisfied

What is Expected Value?
In probability theory (and especially gambling), the expected value (or mathematical expectation) of a random variable is the sum of the probability of each possible outcome of the experiment multiplied by its payoff ("value")

For example, an American roulette wheel has 38 equally possible outcomes. A bet placed on a single number pays 35-to-1 (this means that you are paid 35 times your bet and your bet is returned, so you get 36 times your bet). So the expected value of the profit resulting from a $1 bet on a single number is, considering all 38 possible outcomes:

\left( -$1 \times \frac{37}{38} \right) + \left( $35 \times \frac{1}{38} \right),

What is Random Variable?
A random variable can be thought of as the numeric result of operating a non-deterministic mechanism or performing a non-deterministic experiment to generate a random result. For example, a random variable can be used to describe the process of rolling a fair die and the possible outcomes { 1, 2, 3, 4, 5, 6 }. Another random variable might describe the possible outcomes of picking a random person and measuring his or her height.