Problem Description Recently, we will develop an article's region classifier for localization recommendation.In short, given an article, judge which of the following regions the article belongs to (or does not belong to any region): Dalian, Fuzhou, Chengdu, Chongqing, dongguan, Foshan, Nanjing, Harbin, Hangzhou, Jinan, Qingdao, Xiamen, Shenzhen, Shenyang, Suzhou, Tianjin, Wenzhou, Wuhan, Xi' an, Zhengzhou First google, avoid building wheels repeatedly, but did not find the relevant open source tools

We learned in the previous " Linear Regression" that for training data samples (xi,yi)({x_i},{y_i}), we have the following fitting straight line:
{\widehat y_i} = {\theta _0} + {\theta _1} \bullet {x_i}
We constructed a loss function:
C = \sum\limits_{i = 1}^n {{{({y_i} - {{\widehat y}_i})}^2}}
Indicates the sum of squares of the vertical distances from each training data point (xi,yi)({x_i},{y_i}) to the fitting straight line yi = θ 0+θ 1Xi { \ wide hat y _ I } = { \ theta _ 0 }+{ \ theta _ 1 } \ bullet { x _ I }.

Write in front Karush-Kuhn-Tucker (KKT) condition is an important idea in optimization to solve Lagrange duality problems, and is widely used in operations research, convex and non-convex optimization, machine learning and other fields.In this blog, I try to explain Lagrange duality problem and understanding of KKT condition from the perspective of geometry.If you want to learn more about Convex Optimization, it is recommended that Stanford CS229 and Stephen Boyd's classic

After taking Stanford Andrew NG class and finishing all the exercises with matlab, I felt that I was still not satisfied. Therefore, I decided to use pyton to implement the algorithm in class one by one to deepen my understanding, and at the same time, I also avoided becoming a package adjuster. Ha ha, without saying much, I got to the point.
Kmeans is a classical unsupervised clustering algorithm, and its content is easy to understand.

Recently I learned word2vec and realized a simple word2vec implementation according to my own understanding. The algorithm has a total of three data structures, an input layer, an embedding_lookup layer and an output layer. The simple code is as follows: import tensorflow as tf import numpy as np contents = '一二三四五六七八九十一二三四五六七八九

Recently I saw a blog post about spring's IOC mechanism, which is well understood.However, the original author did not give too many explanations on the positioning of resources, the analysis of resources and the loading strategy of resources in the spring bottom layer, which will be added later.
IOC(DI): In fact, the concept at the core of the Spring architecture is not so complicated, much less obscure as some books describe.

Netty QuickStart The following enumerates the basic building blocks of all Netty applications, including clients and servers.
BOOTSTRAP Netty applications start by setting the bootstrap class, which provides a container for application network layer configuration.
CHANNEL The underlying network transport API must provide interfaces for application I/O operations, such as reading, writing, connecting, binding, etc.For us, this is a structure that almost always becomes a "

1: BN layer
Li Hongyi Video Explanation
bn (batch normalization) layer
(1) Accelerating convergence (2) Controlling over-fitting, which can reduce the insensitivity of the network to initialization weights with little or no Dropout and regularization (3) and allow a larger learning rate to be used.

The zoo have N cats and M dogs, today there are P children visiting the zoo, each child has a like-animal and a dislike-animal, if the child’s like-animal is a cat, then his/hers dislike-animal must be a dog, and vice versa. Now the zoo administrator is removing some animals, if one child’s like-animal is not removed and his/hers