Posts

Showing posts from October, 2011

Uber's Michelangelo vs. Netflix's Metaflow

-
  Uber's Michelangelo vs. Netflix's Metaflow Michelangelo Pain point Without michelangelo, each team at uber that uses ML (that’s all of them - every interaction with the ride or eats app involves ML) would need to build their own data pipelines, feature stores, training clusters, model storage, etc.  It would take each team copious amounts of time to maintain and improve their systems, and common patterns/best practices would be hard to learn.  In addition, the highest priority use cases (business critical, e.g. rider/driver matching) would themselves need to ensure they have enough compute/storage/engineering resources to operate (outages, scale peaks, etc.), which would results in organizational complexity and constant prioritization battles between managers/directors/etc. Solution Michelangelo provides a single platform that makes the most common and most business critical ML use cases simple and intuitive for builders to use, while still allowing self-serve extensibi...
Post a Comment
Read more

Adapter Design Pattern

-
The adapter pattern is an extremely useful pattern for a few reasons. The reason I like it so much is because if you understand it, then you understand the concept of Dependency Inversion (the "D" in SOLID object-oriented design). Dependency Inversion is an object-oriented principle which states that in code, higher level modules should not depend on lower level modules. Instead, lower level modules should depend on higher level ones. Let's say you've got one higher level object that requires a lower level object to do it's job. But the lower-level object is likely to change or be swapped out for another. Here's the bad way to do it: namespace HigherLevelModule { public class HigherLevelObject { ...... public void Run() { LowerLevelObject obj = new LowerLevelObject(); string result = obj.GetSomething(); Console.WriteLine(result); } } } namespace LowerLevelMod...
Post a Comment
Read more

Command Design Pattern

-
Image
The command pattern is best used when you have a command that must be executed by one object, while another object has the responsibility of deciding WHEN that command will be executed. The three entities associated with the command pattern are: client : the object that CREATES the command object and assigns it's receiver receiver : the object that HAS the method that will be run invoker : the object that DECIDES WHEN the command will be run Now let's look at an example. You have an IDE, and the 'undo' command is extremely important. Every action that is performed on the IDE has a method called undo() on it. But the actions have to be stored somewhere after they perform their task though, so that it's possible to call the undo() methods later, right? The solution is to push each action onto a stack when it's run, and register a listener for the 'undo' keyboard-combination/menu-item. The listener will pop() the stack, and invoke the popped ac...
Post a Comment
Read more

Flyweight Design Pattern

-
So you have a group of objects that can be reused, but each time you use one you have to create it first. The problem is that instantiating these types of objects is expensive, and you don't want to instantiate the same kind every time you need it. Enter Flyweight design pattern. With this pattern, the trick is to create a new object only when you've never created that same type of object before - if you HAVE created it before, then simply look it up and use the one that's already instantiated. Here's some code: public class Flyweight { public static Dictionary _Table = new Dictionary (); public ISomething GetSomething(int key) { ISomething something; if(_Table.TryGetValue(key, ref something)) { return something; } else { something = new Something(key); _Table.Add(key, something); return something; } } } What this code does is ensure that t...
Post a Comment
Read more

Factory Design Pattern

-
Image
The factory pattern is used when you want to separate the creation of objects from where they'll be used. For example, say you have one object that must get a Triangle object and draw it on screen. It doesn't care about what the triangle's state is, just that it has a void Draw() method. Suppose further that you have many types of triangles, and they can all be 'drawn' onscreen. In order to separate the creation of triangle objects from the caller, we would use the Factory Design Pattern: 1.) Create an interface called ITriangle, with one method : void Draw() 2.) Have all types of Triangle implement ITriangle. 3.) Create an interface called ITriangleFactory with 1 method: ITriangle GetTriangle() 4.) Create a concrete 'factory' class for each type of Triangle, each of which implements ITriangleFactory, and returns a new instance of their respective triangle in their implementations of GetTriangle()...since each Triangle type implements ITriangle, ...
Post a Comment
Read more