absherwin

C/C++ shouldn't generally be needed. The computationally intensive part is optimized. What makes a given system slow is either poorly optimized numerical code (less common) or doing a bunch of needless repetitive work because the system doesn't make it easy to separate different components of the modeling.

They also won't even be seriously considered at most large companies because the typical person in that role doesn't have the skills and a single-person using a different solution makes it difficult to transition work.

I did a similar analysis several years ago. The two fastest solutions that didn't involve writing lower-level code aren't mentioned: A specialized program targeted at the insurance industry called Emblem and Stata. If I recall, they ran as fast on a single system as RevoScale does on a cluster. Of course, comparing different benchmarks is fraught with peril.

The SAS performance reported is surprisingly bad and likely is a result of the data processing steps involved. A SAS procedure is much more similar to a program than a function. While SAS wasn't the fastest solution, most of its slowness was due to building the design matrix rather than running the actual regression. There are a number of options for both tuning how this is done and caching this preprocessing. The biggest different would be using SSD's on the machine which seems unlikely given the results.

The entire YC team can advise ~66 startups and they have ~14 people which suggests a person can advise 4-5 startups at a YC-level. It's not quite equivalent since YC has the benefit of multiple perspectives though at the cost of increased coordination.

The truth of b & c are completely dependent on who hmexx is.

If this were a one time game you'd be correct. Assuming, he can quickly filter out a significant portion, it's possible to be very involved and have a significant chance of a large exit. This, of course, assumes that he's good at what he does which is an underlying assumption of this whole proposition.

Rationale: Let's assume a home-run rate of about 3% which is my rough estimate of what it is for YC startups. That means that funding 75 startups yields a 90% chance of at least one home run.

Now one might ask how long this will take. If we assume that 50% of the companies can be marked as failures quickly, he will only need to invest substantial time in ~37 companies. If he can focus principally on 3 at a time for a period of 6 months, it will take approximately 6 years to achieve a 90% home run probability. Of course, he'll almost certainly have had some modest successes before then.

Again this is all predicated on hmexx being good at selecting and advising startups. For most people, the odds are much worse.

Better yet use enumerate to save looking up the character index:

    for i,char in enumerate(string):
        print termcolor.colored(char, colors[i%7])

Not quite but closer than I would have suspected:

   Suicide attempt fatality rate: 9%
   Everest attempt fatality rate: 2%

Sources:

    Climber fatality from attempts through 2006:
    http://www.adventurestats.com/tables/EverestAgeFat.shtml

    Suicide estimate from NIH:         
    http://www.nimh.nih.gov/health/publications/suicide-in-the-us-statistics-and-prevention/index.shtml

The key risk for Apple is what its future earnings will be. Given that its current earnings dwarf anyone else, one must believe they will be able to be maintained which is equivalent to a bet that the iPhone will continue to be sold in high quantities with high margins.

In the last year Apple generated $55B in pre-tax earnings. They had $80B in revenue from the iPhone at a margin of >60% or $48B. They sell the iPhone for an average of $642 while the Nexus 4 retails at $299. If the iPhone dropped to a wholesale price of $399 ($100 pricing advantage), it would reduce pre-tax earnings by $30B (55%). If iPhone sales double that would offset about 60% of that decline.

Another way to ask this question is what total mobile phone profits will be in mature market. If there are 7B phones replaced every 4 years, that implies 1.75B phones sold. Cost will likely fall so revenue could be 437.5B. If Apple can capture 30% of that with margins more similar to the Mac (~30%), this suggests even if Apple continues to make better products and can maintain significant market share, they'll earn 20% less from the iPhone than they do today.

Of course Apple has an extraordinarily successful iPad business but its margins are much lower than the iPhone and as an unsubsidized device will likely face greater pressure.

The questions to ask is evaluating Apple are: When will their earnings peak and by how much do will they fall before they reach an equilibrium. While this may be anathema to some and I admire what Apple has achieved, large economic profits cannot exist in the long run in a competitive market. Why is a longer discussion but I challenge the reader to pose a counterexample.

That the sum of two Gaussians is a Gaussian an be found in any basic text on statistics. That combining distributions, calculating them in parallel or online is possible isn't new either. The Kalman filter which was published in 1961 is an example of an online algorithm for fitting a Gaussian.

This is a interesting exposition and places specific results in a more general algebraic framework; however, the title suggests this is a revolutionary discovery which it isn't

A todo list is one function of email. It's also an information sharing and socialization service. The right solution will contain specialized functions for these various uses. Note that a single email can fall into multiple categories. "Prep for 12/3 meeting" might contain information which should be added to a searchable library for future reference and create a to-do entry to ensure that the documents are read before the meeting.

The bigger challenge is that e-mail as a to-do list differs from a calendar in that the time and place of the to-do's is a bit more nebulous. For example, if you send someone an email with a link to your page and request feedback, you have a certain but inexact timeframe in mind. The secret to making this work is providing a heuristic ordering based on the sender's desires and the recipients priorities. The former can likely be specified with a few request classes or duration options. The latter would likely have to be learned from usage history, possibly with a VIP list used as a starting point.

There's no need to assume that a switch for some Macs will be a switch for all Macs. Multi-architecture binaries were standard for several years during the x86 transition. As they move to an App-Store model, it will be relatively simple to download the right binary based on architecture. Signed binaries could be required to be multiplatform. The minority of users who want to install software that's not Apple-approved will have to deal with potentially higher amounts of complexity.

As long as there are a reasonably small number of variables, this could be done scalably. The coworking spaces seem to relieve the space and furnishing concerns but don't fix the HR and legal paperwork headaches.

You mention that by Series A, companies want to customize. Is it more a function of wanting some of their own common space or a desire for individual expression?

Why hasn't one of the incubators or a startup solved these problems? Between apartment phase and some reasonable maturity point (20-50 people?), one could have a reasonably large building or set of nearby buildings that's subdivided from companies of various sizes. It would have the benefit of: Centralized food acquisition either with community eating or distribution to each unit; centralized administrative services; pre-furnished units which simplifies moving; and big company benefits such a onsite gym, dry cleaning, etc.

Presumably such a entity should also provide centralized arrangement of benefits, payroll, legal documents and other issues. An extreme form would be to have this management company run resource companies that make all hires on pass-through basis and submit a single consolidated bill to the main company each month.

Why wouldn't this work? I can think of only three reasons though I'm not convinced any of them are true: Insufficient supply of sufficiently homogenous companies (e.g. software startups); sufficient diversity of preferences for customization, particularly with regard to location to create insufficient density; or fear of poaching due to close proximity to other companies that require the same skillset

What you want to avoid with this sort of AB testing is the same person seeing multiple prices. The easiest, decently close approximation is just to split the page served by geocoded IP. The regions should be split randomly up-front.

Also don't underestimate the value of the more expensive plan to making your $4.95 plan look cheap. A simple answer to what could be premium is contained in your terms: Make the $99/month plan suitable for use in a production environment, as web middleware or a backend, etc by relaxing restrictions and having some sort of SLA.

I agree that Fizzbuzz can be a more interesting example of how to write code without repetition. While the author suggests that languages such as Haskell provide a unique advantage, the deciding question seems to be the availability of pre-built abstractions. Consider the following solution in Python:

  for i in xrange(1,101): print (('' if i%3 else 'Fizz')+('' if i%5 else 'Buzz')) or i

or the even more general:

  mapping={3:'Fizz',5:'Buzz'}
  for i in xrange(1,101): 
    print ''.join(['' if i%x else mapping[x] for x in mapping]) or i

We can even do this in C though to write something as extensible as the second would require writing more helper functions than I can justify for this brief comment:

  #include<string.h>
  #include<stdio.h>
  int main(){
    int i;
    char s[4];
    char n[3];
    for (i=1;i<101;i++){
      sprintf(n,"%d",i)
      s=strcat((i%3)?"":"Fuzz",(i%5)?"":"Buzz")
      printf("%s",(strlen(s))?s:n)
    }
    return 0;
  }