![](https://img-blog.csdnimg.cn/20190918140053667.png?x-oss-process=image/resize,m_fixed,h_224,w_224)
Data-Driven Decision Making
文章平均质量分 95
- Machine Learning
- Reinforcement Learning
- Statistics
- Data driven methods
tangwing
三千年读史,不外功名利禄,九万里悟道,终归诗酒田园
展开
-
Forecasting (二) : Deep Models [ongoing]
Deep modelsDeepARSalinas, D., Flunkert, V., Gasthaus, J., & Januschowski, T. (2020). DeepAR: Probabilistic forecasting with autoregressive recurrent networks. International Journal of Forecasting. https://doi.org/10.1016/j.ijforecast.2019.07.001A.原创 2020-12-20 18:39:32 · 324 阅读 · 0 评论 -
Forecasting (一):introduction
本篇文章希望对demand forecasting涉及的技术进行框架性的整理。首先参考的是供应链及库存相关的著作,一般其中都会有关于forecasting的一章。ReferencesWaters, D. (2003).Inventory control and management 2nd. John Wiley & Sons. (偏OM) Axsäter, S. (2015).Inventory control(Vol. 225) 3rd. Springer. (更新,信息更多些)...原创 2020-11-09 10:49:50 · 531 阅读 · 0 评论 -
MLE、MAP、贝叶斯估计、MCMC、EM
机器学习中的MLE、MAP、贝叶斯估计 - 李文哲的文章 - 知乎 https://zhuanlan.zhihu.com/p/37215276上面这篇文章对所提到的三种方法做了清晰的对比。总结图:另外文中总结:几点重要的Take-aways:每一个模型定义了一个假设空间,一般假设空间都包含无穷的可行解; MLE不考虑先验(prior),MAP和贝叶斯估计则考虑先验(prior); MLE、MAP是选择相对最好的一个模型(point estimation), 贝叶斯方法则是通过观测数据来估原创 2020-10-31 14:04:37 · 1342 阅读 · 0 评论 -
常见分布总结
Source:https://www.youtube.com/watch?v=YXLVjCKVP7U&list=PLTNMv857s9WVzutwxaMb0YZKW7hoveGLS&ab_channel=zedstatistics链接中的Youtube频道对统计相关的各种概念有很好的讲解,强烈推荐。本篇简单摘要记录一下常见的分布。这些常见分布Excel都是支持的。Binomial2 outcomes per trial, proba p or (1-p) Trials a.原创 2020-10-20 10:02:43 · 3793 阅读 · 0 评论 -
[Casual note] Time series prediction
Time series prediction- 在主要feature是时间时用的比较多。如果有很多额外的feature,考虑深度模型- Trend: 长期的趋势- Seasonality:季节性- Cyclicity:周期性 比如经济周期- Irregularity:unpredictable factors。最好在数据层面就剔除掉# Before forecasting- Make sure the series is stationary : devoid of seasonali原创 2020-09-01 10:14:20 · 154 阅读 · 0 评论 -
Multi-agent and distributed RL [Topic under construction]
It's unrealistic to consider the environement as static with only one agent: it's almost always necessary to consider the reaction of multiple agents on each others' moves. With multiple agent, the space of control (action) becomes exponentially large on t原创 2020-08-23 15:50:40 · 146 阅读 · 0 评论 -
Can agents learn inside of their own dreams?
这次阅读一篇NIPS2018的文章,关于World Models in Reinforcement Learning.原创 2020-07-27 16:44:51 · 80 阅读 · 0 评论 -
Pytorch学习纪要[ongoing]
莫烦python教程和视频是不错的入门资源:https://morvanzhou.github.io/搭建神经网络,可以自建class继承torch.nn.Module,也可以使用nn.Sequential快速搭建 使用torch.save(net.state_dict,'file.pkl') 比torch.save(net,'file.pkl')会稍快一些,实测生成的文件前者确实较小。 Batch training: ds=torch.utils.data.TensorDataset(x,y);原创 2020-07-26 15:20:40 · 114 阅读 · 0 评论 -
[NOTE] Advice and Perspectives on RL Research Frontiers - Rich Sutton in DLRLSS 2019
根据我的习惯,当然先放ressources:slides,video. 这是Sutton在DLRLSS 2019 summer school上的一个lecture,从他自己的角度分享了对RL领域的一些理解,他目前的研究方向及前沿等。一些思考还是很有启发的。个别要点摘录于此,细节可以自行阅读、观看。Developing your own research thoughtsThere are no authorities in science. Be ambitious but also humble.原创 2020-07-21 13:48:11 · 139 阅读 · 0 评论 -
[Note In progress] Model-based Reinforcement Learning
Model based methods can be used in Control Theory. Environment has assumptions and approximations.Learn the model. By supervised learning, for instance. Play the game then train the world model. World models: one of my favorite approaches in which th...原创 2020-07-19 14:11:13 · 125 阅读 · 0 评论 -
【转】MCMC采样详解
Check here:https://zhuanlan.zhihu.com/p/37121528转载 2020-07-08 10:30:09 · 296 阅读 · 0 评论 -
[NOTE] Algorithms in E-Commerce Company (Stitch Fix), talk by YuanBo
Stitch FixA DS (Data-Science) driven online shopping model. Customers receives boxes and choose to keep or return the items inside.Culture这个在电商模式上有所创新的美国电商,给予了数据科学非常重要的位置。从组织架构上DS部门的定位不是为其他部门服务,而是提供business insight和innovation. 另外,团队并未像通常一样按照技术智能划分为算法、数据原创 2020-06-21 16:32:04 · 153 阅读 · 0 评论 -
频率学派与贝叶斯学派
贝叶斯学派与频率学派有何不同? - 郭志敏的回答 - 知乎 https://www.zhihu.com/question/20587681/answer/139100761看了一系列资料后我的粗浅层面理解:频率学派(frequentist):认为概率即事件长时间内发生的频率,描述的是客观存在的事实,即涉及到的未知参数其实值是固定的,只是我们不知道。例如,在频率派的哲学语境里,飞机事故的概率指的是长期来看,飞机事故的频率值。对许多事件来说,这样解释概率是有逻辑的,但对某些没有长期频率的事件来说,这样解转载 2020-06-08 07:47:38 · 227 阅读 · 0 评论 -
A Road Map for Deep Learning
点这个:https://towardsdatascience.com/a-road-map-for-deep-learning-b9aee0b2919f原创 2020-06-07 10:37:54 · 139 阅读 · 0 评论 -
[NOTE in progress] ECE236C - Optimization Methods for Large-Scale Systems [on going]
Source:http://www.seas.ucla.edu/~vandenbe/ee236c.htmlIntroductionOutlineFirst-order algorithms Decomposition and splitting Second-order algorithms for unconstrained optimization Interior point for conic optimizationGradientConvexity-∇^2(f..原创 2020-06-06 22:57:26 · 446 阅读 · 0 评论 -
Taylor, Jacobian, Hessian, Newton and all the else about gradient
本文的主要目的是对基于gradient的一些approximation知识点以及优化方法做一个简单的review。详细内容参考引用链接,这里只列出key points,主要是在遗忘的时候能够快速catch up…Jacobian矩阵和Hessian矩阵引用:Jacobian矩阵和Hessian矩阵 雅克比与海森雅可比矩阵(描述的一阶导数矩阵).. 把它理解为一阶gradient就...原创 2020-05-03 11:36:23 · 187 阅读 · 0 评论 -
Brief Intro of Deep Learning【李宏毅课程笔记-待完成】
李宏毅2006, Restricted Boltzmann Machine. Complex. Used to initialize multi-layer perceptron (1980), to be called Deep LearningNetwork -> function setDeep -> many hidden layers2015, Residual N...原创 2020-04-26 16:59:19 · 103 阅读 · 0 评论 -
Stochastic Optimization: Casual Notes
Currently learning stochastic optimization (SO) theory, I will note important content here.Multi-stage vs Multi-periodStage: is defined according to decisions madePeriod: is defined according to t...原创 2020-02-22 08:05:20 · 314 阅读 · 0 评论 -
[Note In progress]Actor-Critic, DDPG and GAN
In the last postOverview of RLwe've seen two different methodologies: Policy Gradient which aims at training a policy (Actor); and Q-Learning which aims at training a state-action value function (Cr...原创 2020-02-16 15:34:18 · 394 阅读 · 0 评论 -
An Casual Overview of Reinforcement Learning
强化学习概览This overview is largely based on this article:https://medium.com/@SmartLabAI/reinforcement-learning-algorithms-an-intuitive-overview-904e2dff5bbc.Model-based vs Model freeModel: world...原创 2020-02-14 22:59:49 · 507 阅读 · 0 评论 -
Activation Functions in Neural Networks
This article is inspired by这里and 这里.- 激活函数的主要意义是为NN加入非线性的元素。在神经学上模仿的是一个神经元是否有效。A Neural Network without Activation function would simply be aLinear regression Model.Neural-Networks are consid...原创 2020-01-31 00:27:41 · 360 阅读 · 0 评论 -
Graph Neural Network: A First Glance
@[TOC]GNNResources从图(Graph)到图卷积(Graph Convolution):漫谈图神经网络模型 (一)VocabularyFixed Point Theorem : a convergency guaranteeContraction MapBP: Almeida-Pineda vs BPTTShort NotesTo make fff a Cont...原创 2019-09-06 17:44:03 · 176 阅读 · 0 评论