Zylophone

System Design Cheatsheet

Picking the right architecture = Picking the right battles + Managing trade-offs

Basic Steps

Clarify and agree on the scope of the system

User cases (description of sequences of events that, taken together, lead to a system doing something useful)
- Who is going to use it?
- How are they going to use it?

Large Scale Distributed Deep Networks

Introduction

In machine learning, accuracy tends to increase with an increase in the number of training examples and number of model parameters.
For large data, training becomes slow on even GPU (due to increase CPU-GPU data transfer).
Solution: Distributed training and inference - DistBelief
Link to paper

DistBelief

Django/Python Questions

Find the most frequent integer in a list, Test list : [3,5,6,3,9,0,3,8,3,1,3]

Find the common elements of 2 lists

    a = [2,3,4,1,1,3,6]
    b = [2,8,9,1,3]

Matroid- Spoke to John. Asked about what projects I've worked with. Told me what Matroid does, how the teams are structured (field deep learning, product, research?), mostly hiring generalists with possible room for one FE specialist. Based in Palo Alto. Seemed interested in moving me forward.

He mentioned they are dealing with new models, doing some product non CV work (i.e. attributed based access control) mentioned the team is using https://github.com/stalniy/casl https://en.wikipedia.org/wiki/Attribute-based_access_control

Great Expectations- Spoke to Martin.

Asked me about my best project. Collecting telemetry data and showing it to Fleet managers.
Describe the software development life cycle. How long did this project take?

GPU Optimization for GameDev

By person

Emil Persson @Humus
- Blog
- <2013> Low-Level Thinking in High-Level Shading Languages
- <2014> Low-Level Shader Optimization for Next-Gen and DX11
- <2018> Rule of optimization
Matt Pettineo @mynameismjp

GPU Optimization for GameDev

By person

Emil Persson @Humus
- Blog
- <2013> Low-Level Thinking in High-Level Shading Languages
- <2014> Low-Level Shader Optimization for Next-Gen and DX11
- <2018> Rule of optimization
Matt Pettineo @mynameismjp

	import heapq
	"""
	At each step, pick two elemens a[i], a[j], i != j and remove them from the array. Add the sum a[i]+a[j] to the end
	of the array. This step costs a[i] + a[j]. Return the minimum total cost to reduce an array to 1 element.
	"""

	def reduceArray(arr):
	total_cost = 0
	arr.sort()
	heapq.heapify(arr)

+## array contains coins represented by 0(head) and 1(tail)
+def minimum_coins(array)
+    head = array.select {|e| e.eql?(0)}.length
+    tail = array.select {|e| e.eql?(1)}.length
+    (array.length/2) - [head, tail].min
+end

	/*****************
	* cellBlockA.js *
	*****************
	*
	* Good morning, Dr. Eval.
	*
	* It wasn't easy, but I've managed to get your computer down
	* to you. This system might be unfamiliar, but the underlying
	* code is still JavaScript. Just like we predicted.
	*

	function functional_scoping() {
	var x = "outside if";
	if(true){
	var x = "inside if";
	}

	console.log(x);
	}

	functional_scoping();