▍1. char-rnn-master
说明: 命名实体识别代码,BiLSTM-CRF代码,Tensorflow框架(Named entity recognition code, BiLSTM-CRF code, Tensorflow framework)
说明: 命名实体识别代码,BiLSTM-CRF代码,Tensorflow框架(Named entity recognition code, BiLSTM-CRF code, Tensorflow framework)
说明: 伯克利课程 dqn神经网络 深度学习 吃豆人游戏(Berkeley courses Neural network dqn Deep learning Pac Man Game)
说明: 完成一个基于q-learning算法的简单寻径实例(A simple path finding example based on Q-learning algorithm is completed)
说明: 用python语言编写的粒子群优化算法,内有多种适应度函数可供选择(Python language used particle swarm optimization algorithm, there are a variety of alternative fitness function)
说明: 基于聚类分析的商圈识别,附上完整版的代码和小论文(Cluster analysis based on the identification of business circle, attached small paper)
说明: 其中x是任何一个大于xmin的数,xmin是X最小的可能值(正数),k是为正的参数。帕累托分布曲线族是由两个数量参数化的:xmin和k。(Where x is any number greater than xmin,xmin is the smallest possible value of x (positive), and k is a positive parameter the Pareto family of distribution curves is parameterized by two quantities :xmin and K)
说明: 各种机器学习方法的源代码,包括决策树、随机森林、神经网络等(Source code of various machine learning methods, including decision tree, random forest, neural network, etc)
说明: 深度学习入门 基于python的理论与实现 包含PDF版本和源码(Introduction to deep learning theory and Implementation Based on Python Contains PDF version and source code)
提升方法是将弱学习算法提升为强学习算法的统计学习方法。在分类学习中,提升方法通过反复修改训练数据的权值分布,构建一系列的基本分类器(弱分类器),并将这些基本分类器线性组合,构成强分类器。提升树是建立在决策树上的一种提升方法。针对回归、分类问题,它采用的损失函数不同。对于回归问题,通常使用平方误差损失函数;而对于分类问题,通常使用指数损失函数。代表性的方法主要有AdaBoost算法以及梯度提升树算法(GBDT)。
说明: 这本书是一本机器学习经典书籍西瓜书的参考书,对其中的公式做了详细的推导。(This book is a machine learning classic book watermelon book reference book, which made a detailed derivation of the formula.)
说明: 图像压缩,人工智能算法,神经网络。python 代码。可以直接使用。(Image compression, artificial intelligence algorithm, neural network. Python code. It can be used directly.)
通过深度学习的方法,能够对图像进行压缩,比传统的方法要好很多,效果更加明显。(Through in-depth learning method, the image can be compressed, which is much better than the traditional method, and the effect is more obvious.)
说明: 通过深度学习的方法,能够对图像进行压缩,比传统的方法要好很多,效果更加明显。(Through in-depth learning method, the image can be compressed, which is much better than the traditional method, and the effect is more obvious.)
说明: 利用强化学习进行环境交互,选择信道,实现无线网络的资源分配(Using reinforcement learning environment interaction, choose channel, wireless network resource allocation)
Python网络数据采集。OReilly.Web.Scraping.with.Python.2015.6,共340页PDF。
说明: 使用多层神经网络对芬兰交通流进行预测。程序包括数据清洗、pandas数据可视化、keras建模、预测结果可视化。数据格式为:点ID,年,天数,小时,分钟,秒,百分之一秒,长度(m),车道,方向,车辆类别,速度(km/h),有缺陷的(0-错误观察,1=错误观察),总时间(技术),间隔(技术),排队(技术)。共16列数据(A multi-layer neural network is used to predict traffic flow in Finland. The program includes data cleaning, pandas data visualization, keras modeling and prediction result visualization. The data format is: point ID, year, days, hours, minutes, seconds, hundredths of a second, length (m), lane, direction, vehicle category, speed (km / h), defective (0-error observation, 1 = wrong observation), total time (Technology), interval (Technology), queuing (Technology). A total of 16 columns of data)
说明: 人工神经网络实验,采用RBF神经网络实现函数的拟合(In the experiment of artificial neural network, RBF neural network is used to realize the function fitting)
说明: 贝叶斯分类器构建网络,对豆瓣进行情感分析;TF-IDF(Using Bayesian classifier to construct network to analyze the emotion of Douban; TF-IDF)