Sunday, March 20, 2016

KASALANAN PALA NI LUCHI AT TV5 EH! #PiliPinasDebates2016… #JHedzWorlD


KASALANAN PALA NI LUCHI AT TV5 EH!


#PiliPinasDebates2016…
KASALANAN PALA NI LUCHI AT TV5 EH!


#PiliPinasDebates2016
http://twitter.com/statuses/711551778651648001


Source by ALDUB CEBU









KASALANAN PALA NI LUCHI AT TV5 EH!

#PiliPinasDebates2016… #JHedzWorlD

VIDEO: Blast off for Europe Russia Mars mission #JHedzWorlD


Europe and Russia have launched a joint mission to the Red Planet.


The satellite, called the ExoMars Trace Gas Orbiter (TGO), lifted off from Baikonur Cosmodrome in Kazakhstan at 09:31 GMT.






VIDEO: Blast off for Europe Russia Mars mission #JHedzWorlD

VIDEO: Blast off for Europe Russia Mars mission #JHedzWorlD


Europe and Russia have launched a joint mission to the Red Planet.


The satellite, called the ExoMars Trace Gas Orbiter (TGO), lifted off from Baikonur Cosmodrome in Kazakhstan at 09:31 GMT.






VIDEO: Blast off for Europe Russia Mars mission #JHedzWorlD

Thursday, March 17, 2016

[Last Samurai?!] 30 Samurai in Modern Armor Gather for Epic Cavalry Fight #JHedzWorlD


On a chilly day in March, samurai amassed once again at an undisclosed location in Tokyo. Are there even more than there were last time?! Just what are they up to?




The truth is the samurai are part of a promotion for an Indiegogo campaign set to begin at the end of March. Tokyo Otaku Mode is fully supporting the launch of this worldwide project by anisong artist and project producer Bamboo of the music group Milktub. The project will be launched soon, but until then you can get the latest information on it here.



This time, the samurai who gathered were all fans of Milktub. As for how they were already wearing Samurai Armor Hoodies, they had prototype models that were available to fans in limited quantities late last year. The fans volunteered their time and gathered together to shoot a promotional video to support the Indiegogo campaign. We thought we’d share with all of you a bit from the actual video shoot. After taking a look you’ll no doubt find yourself wondering what the final video will be like.





First the samurai met up and began their march single file toward the riverbed. They marched in two lines, looking just like a daimyo’s procession.





Of course, they made sure to strictly follow pedestrian laws. There was no need for an all-out charge.





Finally they arrived at the riverbed. First up was a group shoot. Beautiful!
This is where the real photo shoot begins.






The samurai encircled project producer Bamboo like he was a sensei.






The situation then turned more intense. The name of this is “Samurai of War.” Teams are divided into East and West armies and each team charges at once. It looked just like a battle from the Warring States period.





The shoot continued with “Circle of War” as the samurai encircled Bamboo and ran around him.





Next up was the main event: Kibasen. Have you heard of the game kibasen (cavalry fight) that Japanese school kids play? It’s often played as one of the events at athletic meets in elementary and middle school.






Following this, scenes became even crazier than we had anticipated. You probably can’t help but want to watch the video now, right?








Have you ever seen fans actually having this much fun? That’s because the Samurai Armor Hoodies come from a brand that’s loved so much. It’s the wishes of the samurai for this fan community to spread worldwide. Project producer Bamboo is also giving serious thought to holding fan meetings overseas if the campaign is successful. If you want to see modern day samurai all over the world, please support the Indiegogo campaign! And look forward to seeing the completed video too!





★Omake★



The Samurai Armor Hoodie features a removable crest. Fans can even show off their originality with their own unique crest.




#Hairy caterpillar
Themed after military commander Date Shigezane’s hairy caterpillar crest and mustache.



#JSDF
This one features the Kisarazu JSDF crest.



#TOM
There’s even a crest of the TOM logo!



Latest Information:



This is a Tokyo Otaku Mode original article.


Let’s block ads! (Why?)







[Last Samurai?!] 30 Samurai in Modern Armor Gather for Epic Cavalry Fight #JHedzWorlD

Wednesday, March 16, 2016

Sticking with Windows 7? The forecast calls for pain #JHedzWorlD


On Monday morning, I began installing Windows 7 Pro on a clean virtual machine.


On Tuesday morning, more than 24 hours later, the installation was still not finished.


That, in a nutshell, is the unpleasant reality that consumers and small businesses face in sticking with Windows 7 as it counts down to the end of support, 1398 days from today.


The biggest problem, of course, is that Microsoft released Windows 7 Service Pack 1 on February 22, 2011, more than five years ago. Despite occasional rumors and hopes, Microsoft has never released a Service Pack 2 or even a post-SP1 update rollup. Which means that if you try to do a clean install, as I did on Monday, this is what you see when you run Windows Update.


windows7-has-216-updates.jpg


Yes, 216 Important updates are available, and you’d have to be crazy to connect to the Internet without installing those patches first.


But that’s not the worst of it.


Just getting to that screen took more than eight hours, thanks to a bug that was documented last fall in Microsoft Knowledge Base article 3102810: Installing and searching for updates is slow and high CPU usage occurs in Windows 7 and Windows Server 2008 R2.


“Slow” does not even begin to describe the glacial pace at which Windows 7 works before even displaying the list of available updates.


There is a workaround, as I discovered. Download two updates before starting the installation process, KB3083710 and KB3012810, and copy them to a USB flash drive. Install Windows 7 with SP1 and leave the network cable disconnected. Install those two updates from the flash drive before connecting to the network and running Windows Update for the first time, and you can cut that horrendous delay down to a matter of a few minutes.


Even then you have many hours of work ahead of you. When I started over, using that workaround, it still took nearly three and a half hours from the time I started the clean install until those 216 important updates were finished downloading and installing.


And then, after rebooting and checking Windows Update again, there were three more updates. And 26 more after the next restart, and two more after the next restart…


In all, I spent about eight hours to finish installing Windows 7 and something like 300 Important and Optional updates (I lost count, honestly). And each restart along the way demanded my manual intervention.


I could have saved a few hours of that time had I kept a backup from the last time I went through this tedious exercise, in 2013, but there still would have been at least a hundred updates, I expect.


Of course, this problem only affects small businesses and Windows 7 enthusiasts. In enterprise deployments, IT pros build custom images that can be blasted onto PCs using deployment tools. Those are the customers who pay most dearly for the privilege of continuing to run Windows 7, and they have zero interest in simplifying one-off installs.


If I had to point to one feature in Windows 10 that makes it a slam-dunk upgrade over Windows 7, this is it. Updates are cumulative, which means that after a clean install you have to snag only one update (plus a handful of odds and ends, like the latest update for Adobe Flash). Not 216. Or 47. Or even 16


And you can reset a Windows 10 PC–keeping your data files while rebuilding the operating system in place with the latest version, followed by a single update.


Trawling through Microsoft’s Windows 7 support forums, I found more than one commenter who was convinced that the Windows 7 slow update bug is part of a grand conspiracy to make the management process so painful that people will willingly say “Uncle!” and upgrade to Windows 10.


Personally, I think there’s a simpler explanation: Windows 7 is in its “extended support” phase, and Microsoft has moved on. Security updates? Sure. Service packs? Sorry, no.


The pain isn’t deliberate. It’s just the inevitable result of a product whose architecture was designed in 2005 (Windows 7 is, after all, “Vista, fixed”), with no expectation that it would continue to be the dominant operating system for PCs worldwide for more than a decade.


And it’s not going to get any easier between now and January 14, 2020, when the extended support phase finally ends. I wouldn’t be surprised if the update count is up over 350 by then, with a clean install taking days.


Good luck with that.


Let’s block ads! (Why?)







Sticking with Windows 7? The forecast calls for pain #JHedzWorlD

On the edges of the hypercube – codegolf.stackexchange.com #JHedzWorlD



Your job will be to write a function or a program, that will take an integer n>0 as input and output a list of the edges of the n-dimensional hypercube. In graph theory an edge is defined as a 2-tuple of vertices (or corners, if you prefer), that are connected.


Example 1


A 1-dimensional hypercube is a line and features two vertices, which we will call a and b.


enter image description here


Therefore, the output will be:


[[a, b]] 

Example 2


The 4-dimensional hypercube (or tesseract) consists of 32 edges and its graph looks like this


enter image description here


and the output could look like this


[[a, b], [a, c], [a, e], [a, i], [b, d], [b, f], [b, j], [c, d], [c, g], [c, k], [d, h], [d, l], [e, f], [e, g], [e, m], [f, h], [f, n], [g, h], [g, o], [h, p], [i, j], [i, k], [i, m], [j, l], [j, n], [k, l], [k, o], [l, p], [m, n], [m, o], [n, p], [o, p]] 

Rules


  • You may name the vertices any way you like, as long as the name is unique.

  • The edges are undirected, i.e. [a, b] and [b, a] are considered the same edge.

  • Your output must not contain duplicate edges.

  • The output may be in any sensible format.

  • Standard loopholes are forbidden.

Scoring


Shortest code wins.




Python 2, 54 56 62 bytes


lambda n:tuple(k/n,k/n^1<<k%n)for k in range(n<<n) 

Duplicate edges are removed by making a set of sets, except Python requires that set elements are hashable, so they are converted to a tuples. Note that the sets a,b and b,a are equal and convert to the same tuple. xsot saved 2 bytes with n<<n.


This can be cut down to 49 bytes if strings of sets are an OK output format


lambda n:`k/n,k/n^1<<k%n`for k in range(n<<n) 

which gives output like


set(['set([1, 3])', 'set([2, 3])', 'set([0, 2])', 'set([0, 1])']) 


lambda n:[(k/n,k/n^1<<k%n)for k in range(n*2**n)if k/n&1<<k%n] 

First, let’s look at an older version of the solution.


lambda n:[(i,i^2**j)for i in range(2**n)for j in range(n)if i&2**j] 

Each a number in the interval [0,2^n) corresponds to a vertex with coordinates given by its n-bit binary strings. To vertices are adjacent if they differ in a single bit, i.e. if one is obtained from the other by xor-ing a power of 2.


This anonymous function generates all possible edges by taking every vertex and every bit position to flip. To avoid duplicating an edge in both directions, only 1’s are flipped to 0’s.


In the more golfed solution, k is used to encode both i and j via k=n*i+j, from which (i,j) can be extracted as (k/n,k%n). This saves a loop in the comprehension. The powers of 2 are done as 1<< to have the right operator precedence.


An alternative approach of generating each pair of vertices and checking if they are a bit flip apart seems longer (70 bytes):


lambda n:[(i,x)for i in range(2**n)for x in range(i)if(i^x)&(i^x)-1<1] 



JavaScript (SpiderMonkey 30+), 69 bytes


n=>[for(i of Array(n<<n).keys())if(i&(j=1<<(i>>n)))[i&=(1<<n)-1,i^j]] 

This started out as a port of xnor’s Python solution but I was able to save 9 bytes by rewriting the code to use a single loop.




FromLetterNumber/@#&/@EdgeList@HypercubeGraph@#& 

Just an anonymous function that uses built-ins.




2i^:qt!Z~Zltk=XR2#fh 

This works with current version (14.0.0) of the language/compiler.


Try it online!


Explanation


This uses more or less the same idea as @xnor’s answer.


2i^ % take input n and compute 2^n :q % range [0,1,...,2^n-1] (row vector) t! % duplicate, transpose into a column vector Z~ % bitwise XOR with broadcast Zl % binary logarithm tk % duplicate and round down = % true if equal, i.e. for powers of 2 XR % upper triangular part, above diagonal 2#f % row and index columns of nonzero values h % concatenate vertically 



ạ/S=1 2Rṗœc2ÇÐf 

Try it here. For input 3, the output is:


[[[1, 1, 1], [1, 1, 2]], [[1, 1, 1], [1, 2, 1]], [[1, 1, 1], [2, 1, 1]], [[1, 1, 2], [1, 2, 2]], [[1, 1, 2], [2, 1, 2]], [[1, 2, 1], [1, 2, 2]], [[1, 2, 1], [2, 2, 1]], [[1, 2, 2], [2, 2, 2]], [[2, 1, 1], [2, 1, 2]], [[2, 1, 1], [2, 2, 1]], [[2, 1, 2], [2, 2, 2]], [[2, 2, 1], [2, 2, 2]]] 

I hope [1, 1, 1] etc. is an okay “name”.


Explanation


The first line is a predicate on a pair of edges: [A, B] ạ/S=1 is equal to sum(abs(A - B)) == 1, i.e. A and B differ in exactly one coordinate.


The second line generates all edges with 2Rṗ (cartesian power of [1, 2]), then all pairs using œc2 (combinations of size two without replacement), then filter (Ðf) by the above predicate (Ç).




fq1.aT.c^U2Q2 

Output on input 3:


[[[0, 0, 0], [0, 0, 1]], [[0, 0, 0], [0, 1, 0]], [[0, 0, 0], [1, 0, 0]], [[0, 0, 1], [0, 1, 1]], [[0, 0, 1], [1, 0, 1]], [[0, 1, 0], [0, 1, 1]], [[0, 1, 0], [1, 1, 0]], [[0, 1, 1], [1, 1, 1]], [[1, 0, 0], [1, 0, 1]], [[1, 0, 0], [1, 1, 0]], [[1, 0, 1], [1, 1, 1]], [[1, 1, 0], [1, 1, 1]]] 

Explanation:


fq1.aT.c^U2Q2 Implicit: input = Q ^U2Q All Q entry lists made from [0, 1]. .c 2 All 2 element combinations of them. f Filter them on .aT The length of the vector q1 Equaling 1. 


JHedzWorlD







On the edges of the hypercube – codegolf.stackexchange.com #JHedzWorlD

Maine Mendoza scent for local lifestyle brand, sold out in 15 minutes #JHedzWorlD

Maine Mendoza scent for local lifestyle brand, sold out in 15 minutes #JHedzWorlD




Maine Mendoza scent for local lifestyle brand, sold out in 15 minutes #JHedzWorlD