rmenezes

Because I can.

[rod333]

Will lectures on stages still exist?

Princeton's new president, Chris Eisgruber, said that video lectures will never replace lectures on stage:

Eisgruber at @TEDxPrincetonU: Video lectures will never replace "lectures on stages" | http://t.co/AaSg5ej7KS #MOOCs

— Daily Princetonian (@princetonian) November 25, 2013

Meanwhile, Stanford president John Hennessey said that we would move away from lectures and towards more tutoring and collaboration in the classroom.

I like Eisgruber a lot, but I have to side with Hennessey here. I've had fantastic lectures at Princeton, but I've also had terrible lectures, where the majority of students skipped half of the class. To combat this, some professors tried making a "participation grade" and counted attendance in class. I've had lectures so bad that students were willing to take a 5% hit to skip!

One solution is to get better lecturers, but I think the immediately actionable solution is to do more precepts! Princeton has a fantastic precept system, where 20 person classrooms can ask teacher questions, learn from each other and collaborate on homework assignments in class! I've rarely heard complaints about boring precepts.

Consumption of raw knowledge happens more easily at people's own pace - in a book or website. The real question is: how can textbooks need to be improved? I've had plenty of times where I read a proof in a textbook or a professor is teaching an example problem where I get lost - what was that variable again? What does that notation mean?

Bret Victor showed us that programming is more fun with instant feedback, when we can tinker with variables and inspect them at any point in the process. I wonder, in physics and mathematics education, how else can we use instant feedback? Imagine a problem where all the logical steps are explained out. If you don't understand one of them, you can click on a step and expand it - you could keep going down all the way to the theory and get explanations right there.

I think the lectures are going to become endangered. There are more effective ways to transfer raw information to students.

[rod333]

Electrical Engineers will inherit the earth

There's a growing buzzword in the software industry - "designgineers".

"Desingineer" -> mythical person startups are looking for who can do UI, UX and also excellent front- and back-end coding.

— chris dixon (@cdixon) December 19, 2011

It is often said that these designer/developer hybrids are the most sought-after people in the industry. These people can make an entire SaaS product by themselves, make it look good and make it run well. However, I think the future belongs to hardware/software engineer hybrids, which I will call Wozniaks.

Recently, I read iWoz, Steve Wozniak's excellent autobiography. I realized, when reading it, that it is very difficult to find anyone nowadays with the intimate knowledge of both software and hardware that Woz has. Heck, it's already hard to find a software engineer that knows the software half of what Woz needed to do the Apple I and II - system programming, operating systems, graphics, assembly, drivers.

Those skills are going to become more important again. With 3D printing and other innovations, hardware will inevitably become as easy as software. We're already hearing about people starting hardware businesses in their garages and I don't think it's far-fetched to think that the next Nest could come from a person working alone, tinkering with an Arduino and a 3D printer.

Someday, there's going to be an AWS for Hardware - a step past 3D printing as a service. You'll upload blueprints and this service will take orders, do distribution and scale with your business. A factory factory, if you will, that will introduce new products in industries we've never even dreamed of!

Software is easy right now, which is one of the reasons why good design is important to distance yourself from competitors. But when this new hardware era comes, those electrical engineer/software engineer hybrids are going to be the most desired recruits in engineering teams. Bonus points if they can do industrial design too!

[rod333]

When Carl Icahn Tweets, 17 billion dollars appears

Had a nice conversation with Tim Cook today. Discussed my opinion that a larger buyback should be done now. We plan to speak again shortly.

— Carl Icahn (@Carl_C_Icahn) August 13, 2013

After this tweet from Carl Icahn in August, Apple's stock moved up 4%, increasing their market cap by 17 billion dollars. Yowza.

[rod333]

The end of the dot com era

Recently, a lot of people have been criticizing ICANN's new move to open up gTLDs. Some have accused the plan as "serving corporations over individuals."

From a purely economic sense, nothing could be further from the truth. Here are the requirements for a good domain system:

Reliability. Applicants to gTLDs need to prove that they will be good curators of their domain. new gTLD operators will own a piece of the internet. Yes, there absolutely needs to be a system to show that someone will take care of it. We can't expect someone to operate a gTLD from their garage.

Financial stability. Guess what? ICANN needs money, too. These gTLDs ain't gonna be cheap. They need the money to:

Afford the legal challenge. 60,000 of the 185,000 dollar fee is set aside for lawyers.

Operations. They also need money for the increased evaluators, judging panels, increased staff.

Help others in need. They will use some of the money to subsidize the cost for "needy" applicants in developing nations.

Disincentivizing squatters. gTLDs shouldn't suffer the same problem as domains and making them expensive will reduce that likelihood.

Respecting trademarks. ICANN is purposefully giving a "sunrise" period for existing trademark owners to get their domains. That way, they can't be extorted. Makes sense.

In my opinion, the biggest benefit of the new system is that the new real estate will make domain squatters less powerful. By moving away from only 22 gTLDs, we'll finally see the end of ".com" being the king of TLDs. Getting a good domain name has become a hassle and new TLDs like .app will make these brokers less powerful.

[rod333]

Uploading Django's static/media files from Heroku to Amazon S3

A lot of people have trouble deploying Django to Heroku and uploading static files to amazon s3. I've had this problem before, it's a terrible problem to have, so here's a brief guide to help you through it. As usual, the Django community is pretty awesome, so there's already a solution to the hard part of this problem: it's called [django-storages](http://django-storages.readthedocs.org/en/latest/index.html). Essentially, django-storages allows you to plug in the storage backend you want and will allow Django to automatically upload it when you call 'python manage.py collectstatic'. At this point, Heroku automatically calls `manage.py collectstatic` when you push code, so you won't ever need to worry about statics or media files again once you do this correctly. To start, pip install django-storages: pip install django-storages And add storages to your settings.py file INSTALLED_APPS = ( ... 'storages', ... ) Then, in your settings files, add the necessary parameters for S3 if you're in production: if PRODUCTION: ... other stuff ... AWS_STORAGE_BUCKET_NAME = 'yourapp' S3_URL = 'http://yourapp.s3.amazonaws.com/' # Personally, I like to make another S3 bucket for development so I can keep the media files away from my git repo and so other people can use the same media files when they develop. else: AWS_STORAGE_BUCKET_NAME = 'yourappdev' S3_URL = 'http://yourappdev.s3.amazonaws.com/' Make sure your media url is correct: MEDIA_URL = S3_URL + 'media/' Make sure your static urls are correct: if PRODUCTION: STATIC_URL = S3_URL + 'static/' else: STATIC_URL = '/static/' Add the locations of static files into STATICFILES_DIRS (this allows django to automatically find your static files when you call `python manage.py collectstatic`) STATICFILES_DIRS = ( CURRENT_DIR + "/static", ) Make sure you have your STATICFILES_FINDERS: # List of finder classes that know how to find static files in # various locations. STATICFILES_FINDERS = ( 'django.contrib.staticfiles.finders.FileSystemFinder', 'django.contrib.staticfiles.finders.AppDirectoriesFinder', # 'django.contrib.staticfiles.finders.DefaultStorageFinder', ) Now the fun part - the other parameters for storages: DEFAULT_FILE_STORAGE = 'yourapp.s3utils.MediaRootS3BotoStorage' STATICFILES_STORAGE = 'yourapp.s3utils.StaticRootS3BotoStorage' AWS_ACCESS_KEY_ID = 'blahblahblah' AWS_SECRET_ACCESS_KEY = 'blahblahblahblah' AWS_PRELOAD_METADATA = True # forgot what this is for, to be honest from S3 import CallingFormat AWS_CALLING_FORMAT = CallingFormat.SUBDOMAIN This is a custom storage backend that I took from [stack overflow](http://stackoverflow.com/questions/10390244/how-to-set-up-a-django-project-with-django-storages-and-amazon-s3-but-with-diff). It allows you to keep your media files in one folder in S3 and your static files in another folder in S3. Not entirely sure why I have the modified_time function in there - it was a while ago. I think it was solving some S3 datetime bug that was bothering me. # s3utils.py from storages.backends.s3boto import S3BotoStorage from django.utils.functional import SimpleLazyObject from boto.utils import parse_ts class ModifiedS3BotoStorage(S3BotoStorage): def modified_time(self, name): name = self._normalize_name(self._clean_name(name)) entry = self.entries.get(name) if entry is None: entry = self.bucket.get_key(self._encode_name(name)) # Parse the last_modified string to a local datetime object. return parse_ts(entry.last_modified) StaticRootS3BotoStorage = lambda: ModifiedS3BotoStorage(location='static') MediaRootS3BotoStorage = lambda: ModifiedS3BotoStorage(location='media')` Still with me? Sweet. You need amazon s3's python api code in there somewhere. I wasn't sure what the right solution to that was, so I just added it to my project. You can find the code [here](http://codr.cc/5b1964). I _think_ that's it. If you have any questions, feel free to email me at [email protected]

[rod333]

What's the best icon for notifications?

Github's notification icon puzzles me:  It's a dot! Odd, right? I don't like it for a couple reasons: * Even when it's active, I don't notice it. * It has no sense of urgency - I'm not compelled to click on it because it doesn't seem that important. * No one associates a dot with notifications. But that raises the question: what would be a better icon for notifications? After some research, I'm surprised to learn that there are a large variety of icons used for notifications. Here's a small sample: ### The Bullhorn  Used by Spotify and is commonly used in many web services. It's becoming more common every day because it's included in the widely used FontAwesome project. The bullhorn isn't necessarily the first object I associate with notifications, but it definitely has some benefits; it has a sense of urgency and seems to say "listen up!" to the user. ### The Signpost   I'm pretty sure this was the old Facebook notification icon. At first, I thought I associated notifications with sign posts the most. However, I think it's the little number in the red background that gets me, which bring me to: ### The Globe  Facebook's current icon. Maybe the idea here is "the world is trying to contact you"? I think that most people don't notice the globe because they only care about the little number in the corner. Red's a great color to attract attention and it pops out enough to encourage the user to click on it, but not enough that it distracts from the rest of the page. ### No icon, just a number This is probably the oldest way to do it - it's been around since MySpace!  When it's done right, I think it's also the most tasteful method of representing notifications:  At this point (thanks to Facebook) people associate notifications with white numbers on red backgrounds. The red is the most eye catching part of Facebook's top bar. I think this is a great example of G+ design emphasizing the most important part of that button: the number of notifications you have. At the end of the day, that's the most important part about the icon.  You can argue that the zero may not be useful for users looking for old notifications. However, I think it's pretty obvious to regular users of G+, who are the people who would be looking for a list of old notifications anyway. Personally, I like the number more than the bare bullhorn, globe and signpost (and I definitely like it more than the dot!).

[rod333]

Practice doesn't always make perfect.

I played tennis for ten years. I went to practice countless times, had tennis lessons multiple times a week and grew up watching tournaments on television. The only thing I hated about tennis? I sucked at it. I sucked because I didn't do the real thing often enough. I ran a lot of drills but at the end of the day, the only thing the drills taught me was __how to be good at drills__. In practice, I played as well as players would beat me in a match 6-0 6-0 later that day. I never learned how to think during a match, how to play the actual game of tennis. Worse, practice gave me the terribly high expectation that I would do as well in a game as I did in practice. I did so badly in tournaments that I was often afraid to sign up to other ones, creating a negative feedback loop - I would get better at practice, grow higher expectations for tournaments, do badly at tournaments and increase my desire to practice. The 'practice loop': a desire to to go back to practice because you're good at practice, but not the real thing. ### The practice loop in education The same thinking applies to other areas - like computer science education. In Princeton, we keep learning about __theory__ in Computer Science. We go over the theory so many times that we get toasted when we go to the real world for the first time. At my internship, I was left in awe by other interns who major in "software engineering" instead of "computer science". Plenty of computer science majors from Princeton look at these kids who've been programming since the age of six and shrug their shoulders. They say "I'll get better eventually" and they avoid making things outside of class. Wrong - that kind of experience doesn't come from theory classes. The solution is to take the initiative to learn, even if progress will be excruciatingly slow and disheartening. If you put it off and say "oh, I'll focus on getting a good GPA in school", all you'll be doing is getting good at practice. It's also one of the reason I don't sweat about my grades that much. I care very little about getting an 'A' in a class. With 150 really smart people in the class, it takes considerable effort to go from a 'B+' to an 'A'. After a certain point, the marginal benefit in understanding every single intricacy for an exam is very low - I'm not going to remember tiny details ten years from now. ### Practice is good, but not a cure-all Does this mean that practice is a bad thing? Absolutely not. Practice teaches you fundamentals. Just as it is beneficial to perfect your form in tennis, it's very beneficial to learn about the fundamentals of probability and algorithms in Computer Science. I would be legitimately distressed if I do not understand the material in class. But practice is _not a replacement for the real thing_ and I wish I knew that when I was in high school, struggling to get on the varsity tennis team.

[rod333]

What if Django was more like Meteor?

After watching [Meteor's](meteor.com) video for the first time, I was blown away by the idea that you can use server side code the on the client. The consequence of this is that you don't have to create AJAX endpoints on the server to access db information. Soon, I started thinking "man, I wish I could do that in Django." Django's generic views are great, but they're no replacement for accessing the ORM itself. I thought about it some more and then it hit me: Python's syntax is nearly identical to JS syntax when it comes to objects - it might be feasible to fake using Django's QuerySets on the client. So what if Django was more like Meteor? Let's wireframe a Django app called Burt, that would do this for us: ## STEP 1, INSTALLATION: Add this to INSTALLED_APPS: INSTALLED_APPS = ( ..., 'burt', ... ) And this to urls: urls = ( ... (r'^burt/', include('burt.urls')), ... ) ## STEP 2, JS INTEGRATION: Include you would include this into your html: ## STEP 3, PROFIT: In the beginning of your template, you define any permissions you want your client to have in a template tag: {% burt 'appname' 'modelname' get %} {% burt 'appname' 'modelname' edit %} {% burt 'appname' 'modelname' create %} {% burt 'appname' 'modelname' delete %} Upon compiling the template, Burt would allow you to use JS like a Django queryset through a global variable called `burt`: burt.objects.get('appname', 'modelname', { variables here }, boolean autoupdate ); // returns burtObj burt.objects.filter('appname', 'modelname', { variables here }, boolean autoupdate ); // Returns an array of burtObj These functions would use AJAX calls to return the results of the equivalent Django ORM query. As Python syntax and JavaScript syntax are nearly identical when it comes to objects, it would be feasible for the results of these queries to be JS objects that contain nearly the same fields and methods as the equivalent Django models. This would even allow you to access Python meta methods like \_\_unicode\_\_() and Django built-in methods like delete() by making AJAX requests to the server. Burt would do all the work in creating these JavaScript objects and in creating the necessary Django infrastructure. Even better? Burt could take an 'autoupdate' variable that, when set to true, would upload changes to the server automatically. It would be simple: if set to true, the object would be appended to a list in Burt.js. Burt.js would then have an interval that runs and checks a list of objects to see if they are updated, and if so, make an AJAX request that will upload the changes. That means that if you did this: {% burt 'appname' 'modelname' get %} var myObj = burt.objects.get('appname', 'modelname', { id = 0 }, True); myObj.name = "Santa Claus"; Burt would automatically edit myObj in the db upon changing the name. This would make Django have some of Meteor's best features - client side code that emulates the server and autoupdating. Why is it named Burt? Because my desktop wallpaper is a picture of Burt and Ernie and it was the first thing I thought of :) [Tell me](mailto:[email protected]) if you would use something like this. I totally would.

[rod333]

BOINC on the web browser

As the personal computer becomes more potent, the potential benefit of massively distributed computing projects increase. After all, you have more and more users with $2000 facebook machines whose spare computational time could be used on protein folding or SETI. However, the potency of current volunteer computing projects are limited by, well, the number of volunteers.

That's why distributed computing on the web browser is such an interesting topic. Theoretically, through a hot-linked <script> tag or even an <iframe>, a website owner can introduce JS code that will allow any visitor to a page to contribute their computing time and send their results. Previous studies have shown that this is at least feasible.

Although implementation and ethics raise some very compelling issues on their own (which I will address in a later post), there's an even more important question to consider: how would you incentivize web site owners to link to these computing projects?

The answer: a single, reputable distributor of JS for volunteer projects. A BOINC for the browser, if you will. I've been exploring this topic for a paper under the guidance of Professor Brian Kernighan last year in Princeton. So let's look at what this platform could look like on a high level.

BOINC@web

Back in 2002, some very smart people realized that volunteer distributed computing projects had trouble finding the attention necessary to do meaningful amounts of computation. They created BOINC to be a platform for these projects, so that users can discover new projects and developers could have an easy API for development.

Suppose we wanted something similar - a service that reviews and approves code from academics that can be distributed via iFrames:

When linked to the work URL, the service would assign visitors a random project so that each participating project gets computational time. As each project needs a large amount of time to produce meaningful amounts of data, this vistor would continue to work on the same project whenever linked to the same link and would either upload results to the server to cache them locally.

However, some websites wouldn't like the idea of contributing to a project like SETI. Some sites would prefer to contribute to medical simulations. So why don't we let participating websites to choose what causes to contribute to. Personalization would definitely increase the incentive for web sites to contribute. This service would take note of who referred the visitor, check to see if the referrer has any preferences and assign jobs accordingly. Even better: some participants could visit an an XAuth-like url and save their preferences across domains (or opt-out of the service entirely).

It's inefficient that we have a supply of unused computation and increasing demand. I hope we'll address this issue one day, one way or another. For my own studies, I've created a rough version of what this service could look like. When linked to http://jp.rmenez.es/work, a user is randomly assigned one of two mock projects. If they change their preferences at the home page, every future link to that work url will have them working on their preferred project.

Note: http://jp.rmenez.es has been taken down because I'm switching my hosting right now!

The implementation details are quite interesting and the code can be found on Github.

Extra credit: an alternative to advertisements?

There are certain industries that would pay for heavy computation. One non-profit paid over four and a half thousand dollars on Amazon’s Elastic Compute Cloud to create a 1,500 core cluster for three hours for cancer research. Pharmaceutical companies may want to pay for medical simulations or banks may want to pay for a lot of number crunching. Who knows what the future of heavy computation holds? Maybe a browser implementation of sufficient size would be more effective than EC2. If that's ever true, selling computation at this level could be profitable for the organizer of this central service and for website owners.

What would visitors get? Maybe an ad-free experience. I'd take that.

[rod333]

Google CDN Backup

This is pretty cool. Michael Wright on Forrst posted this backup for jQuery loaded from Google’s CDN:

[rod333]

Tech Debt Time Bomb

I’m currently working on a big project that will require thousands of lines of code and a small part of my project requires me to efficiently implement a Fast Fourier Transform in JavaScript. I want to meet my deadline and I’m sure someone’s done this before, so I google it and I use what I find.

And hey, it works! It’s doing exactly what I want! I keep going and adding new features to my project and fleshing it out until I find bugs. A lot of bugs. An unholy amount of bugs. Little did I know, the code that I introduced didn’t work in case X or in scenario Y and I only find out weeks into the development process.

I call this a tech debt time bomb. Time keeps ticking until that one painful, horrifying, oh-so-humbling moment where you realize, after hours of debugging, that the problem is something that you never tested properly in the first place.

Tech debt time bombs are the worst kind of technical debt - a lack of testing. Acceptable technical debt is a conscious decision to not complete a project to your standards and to move on to something else. Sure, if you had more time, you would have changed the structure of X, you have refactored Y, etc. However, you decide what you have is good enough for now and you want to move onto something else before you improve existing code. You’re saying the marginal benefit of moving on is greater than than the marginal benefit of improving your code due to time constraints.

Under no circumstances should you wait to do more tests. As time goes by, your familiarity of the code decreases and existing bugs will cost more man-hours to find and debug. Unless you consider that as part of your marginal cost (and let's face it, no one does), you should not move on without adequate testing.

That's why tech debt time bombs are so terrible. Two months from now, you’re not going to know where all these bugs came from and it's gonna suck.

So look at the code of the open software you’re using. Test it. A half hour now will save you three hours later.

[rod333]

First Post

I'm trying to see if I like Tumblr more than Posterous.  So far, it seems like Tumblr is prettier, but I'm frustrated that someone already took "rmenezes.tumblr.com".

At this point, I've thrown away the idea of making my own blogging platform or hosting it on my own server. Too much of a hassle and I'm not sure if my server could carry the necessary load if I actually say something interesting.