Matrix Review

Matrix Review

Progress/FeaturesSource: Workshop Notes, Matrix Sketches,

Trello, GXT Grid examples, Ideas…

Progress/Requirements

Demo

• http://128.196.105.79:8080/mr/

Progress/Requirements

?

?

• Control Mode: Numerical to specify unit?• Or extra feature to convert values between units?• How to store values in backend? Retain original e.g.

“3mm” “4 cm” before converting anything?

Progress/Requirements?

Progress/Requirements• Additional features:

– Filtering of character values (default: free text)– Control mode: Numerical, Categorical, Off

• Control mode controls filtering options: free text, checkbox, > < =– Commenting values, taxa, characters– Lock values, taxa, characters, grid– Coloring– Hovering values:– Hovering shows a “summary”

• matrix control cell• Taxa• character

– Frozen character row, taxa column– Drag and Drop move of taxa and characters

• Includes autoscroll when dragging next to boundaries of scrollable area

Character A of Organ B of Taxon C has value XColor red “not sure”Comment: “Can you take a look?”

Decisions made along the way

• No character – organ grouping but rather “name based”– Will make character reordering complicated

• Categorical control input box not to add new values: Possibly introduces noise. New values can be added by switching between modes

• No word wrap cells: Makes matrix less space efficient and values can be shown in their entirety in tooltip

• “Grouping grid” / “Filtering” / “Hiding” of taxa– Leverage taxonomic hierarchy: TODO

Feature Possiblities• Taxa ordering mode: Taxonomic hierarchy

– Taxa carry hierachical information and rank in model • A tree• Order taxa (rows) according to DFS of tree (+ Indentation?)• Color code taxon cells for ranks? (conflict with annotation coloring: Color depending on a mode?)

• Hierarchy view to manipulatehierachy for desired taxa order• Hierarchy allows to filter descendants

familygenusspeciesspeciesgenusspecies

Feature Possiblities• Taxa ordering mode: Taxonomic hierarchy

– Alterative: Try to marry table with tree– Expansion of tree node shows additional rows for the child nodes– Probably complex to implement

• Other ideas?

Feature Possiblities

• Switch between taxon order modes or a one-time-action– Complete manual ordering via Drag and Drop and

possibly a new ordering view– Alphabetical– Coverage– Hierachy

• E.g.– Rename a taxa in alphabetical mode: Is it moving or

staying until alphabetical order is triggered?

Feature Possiblities

• Analysis panel (e.g. to slide in from bottom)– Allows to edit matrix while “analyzing”– http://www.sencha.com/examples/#ExamplePlace

:aggregationgrid– http://www.sencha.com/examples/#ExamplePlace

:livegroupsummary– At some point: Diagrams, etc.– Compare open requirements shown previously

Feature Possiblities

• Row expander – http://www.sencha.com/examples/#ExamplePlace:rowex

pandergrid– Would this be a prefered way to show the description?– Configurable to not span over entire row? (description

not well readable)– DnD of values should be possible (Click in character

values table no longer reserved)

• Other options to display description if needed? – Cf. Matrix sketch

Feature Possiblities

• Preview step for entering controlled mode of a character– Numerical

• Specify unit• Range values?• Are there Min/Max allowed values?• …?

– Categorical• Overview of states currently in table: Can review by inspecting

examples where they appear• User can manipulate the list• Upon confirmation, possibly values in matrix are removed

Open Questions• Visualization

– Color code: Multiple aspects, how to do?– Aspects I can currently see

• Colored e.g. “Expert can you verify?”, • Colored e.g. “interesting”• Touched• Visited• Locked• Merged• Dirty• Commented• Rank• Control Mode• Taxon name• Character name• Organ

– Ways to incorporate information: • Cell Tooltip• Cell Color• Cell Background icon(s)• …?

Open Questions• Selection mode

– Cell vs. row: Click is now available in character values grid– Single vs. multiple? E.g. for some kind of compare functionality?

• Column Autosize not supported

• Small tweaks– Control mode: Look ahead to filter values? Column seperation lines– Stripe rows– Track mouse over

• Widgets where appropriate:– Progress bar instead of Coverage <div>?– Color cell?– Date?– Taxon name: Button to open text?– … that’s where the user is supposed to dream– http://www.sencha.com/examples/#ExamplePlace:cellgrid

Progress/Requirements

Parallel Computing

• Dining philosophers problem

Synchronization

• “Doing things at the same time trying not to get into each others way“

• Doing things at the same time can mean many things, crucial for us is: sharing computational resources, e.g. stanford parser access

Another way of looking at it…

You as the designer/coder are “creating the streets and traffic rules”

Critical Section• The critical section (CS) of a process is the block of code

where shared state (e.g., global variables) are accessed and possibly manipulated– (Sidenote: Knowing this should influence the design of your code)

• Problems:– Non-deterministic execution order: Explosion of possible behavior– Data Race, Race Condition– Mutual Exclusion– Absence of Deadlock– Absence of Starvation

public class ApplePicker() {

private int count = 0;

public void pickApple() { count++; System.out.println(“Picked an apple, we now have: “ + count + “apples”); }}

Synchronization Primitives

• Semaphore, lock, monitor, barrier, or latch to guard critical section

• Block execution until “free”. Don’t do:

boolean resourceIsInUse = true;…while(resourceIsInUse() { }

Example

• Shared resource by multiple threads: HashMap -> ConcurrentModificationException

• http://docs.oracle.com/javase/7/docs/api/java/util/HashMap.html

Synchronization in Java• Each object in java has a lock associated• The thread who holds the lock may perform actions on the object, others have

to wait.

• Also, each object Is associated with a monitor: wait(), wait(n), notify(), notifyAll() can be used to synchronize

• If interested in details, e.g. http://pages.cs.wisc.edu/~fischer/cs538.s06/lectures/Lecture34.4up.pdf

public class ApplePicker() {

private int count = 0;

public synchronized void pickApple() { count++; System.out.println(“Picked an apple, we now have: “ + count + “apples”); }}

Object obj;….synchronized(obj) { //implicitely getLock(obj) obj.doSomething(); //implicitely freeLock(obj)}….

In practice

• Speed up of computation -> Optional concurrency to obtain result quicker Vs.• “Multiple agents” – like application-> Inherently concurrent execution

E.g. POS-Tag sentences vs. modeling Dining Philosophers

In practice• Identify part of code that can naturally be parallelized (can be

executed independent from each other, e.g. sentence POS tagging)• Then likely no critical section at all • Avoids all complications and can be implemented relatively easy.• However, beware:

1. Sentence POS tagging itself may be in nature independently solvable for each sentence. However, the resources to complete the task (e.g. Stanford CoreNLP) may pose a CS. They then can either be instantiated multiple times, synchronized as shown, … or rewritten

2. Not all parallelization results in speedup. a) Depending on the task to solve the overhead to create and manage a number

of threads can overweight.b) Amdahls law

In practice

• Debugging becomes harder, due to • Non-deterministic execution order: Explosion of possible

behavior• What that means• Multiple threads in Eclipse debug mode that may need to

be stepped through and coordinated to achieve the desired execution order

• Log files are written by multiple threads at the same time leading to less intuitively readable results

• Turn off parallelization for debugging, if possible

• As an addition: I believe all of the classes we use from Java for creation of threads and to synchronize them if necessary * come from this package:

• http://docs.oracle.com/javase/6/docs/api/java/util/concurrent/package-frame.html

• Notable ones (because they are used in our code):• ExecutorService

• Future

• CountDownLatch • and subclasses of those