Welcome to dl Documentation

Documentation for dl.

Quick Links

• 
• 
• 
• 
• 

Table of Content

C1 - Neural Networks and Deep Learning

Table of Content

Week1 - Introduction to Deep Learning

Table of Content

What is a neural network?

Size - Neuron -> Price

Concept

ReLU Function

ReUL (Rectified, Linear, Unit) is a function that start with zero, and going up linearly.

Week2 - Logistic Regression as a Neural Network

Table of Content

Binary Classification

Logistic Regression

逻辑回归是用于二值分类的情景，二值分类时我们更多的是想知道输出为1的概率值，通常情况下大于50%我们就认为是1，小于50%就认为是0。对比线性回归，用一个向量积+一个常数得到的结果通常是一个数值，可能是负，也可能非常大，这个结果很不适合用来表示概率。

所以逻辑回归的本质就是在线性回归的基础上再加上一个sigma函数，将 (-∞, +∞) 这个区间映射到 (0, 1)这个区间上。

                              1
σ(wT * x + b) = σ(z) = ---------------
                        1 + e ** (-z)

        | /------
        |/
   0.5  /
       /|
------/ |

Logistic Regression Cost Function

传统的LMS误差对于二值分类来说不是一个好的Cost Function，因为会导致Cost Function Non Convex。

逻辑回归的Cost Function长这个样子：Cost(y', y) = - ( y * log(y') + (1 - y) * log(1 - y') )

我们的参数优化目标就是Minimize Cost Function的值。

About the Author

(\ (\
( -.-)o    I am a lovely Rabbit!
o_(")(")

Sanhe Hu is a very active Python Developer Since 2010. Now working at Whiskerlabs as a Data Scientist. Research area includes Machine Learning, Big Data Infrastructure, Block Chain, Business Intelligent, Open Cloud, Distribute System. Love photography, vocal, outdoor, arts, game, and also the best Python.

• My Github: https://github.com/MacHu-GWU
• My HomePage: http://www.sanhehu.org/

API Document

• by Name
• by Structure