<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[todingdong的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/todingdong</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; todingdong]]></copyright><item><title><![CDATA[课整理Seq2seq、attention、transformer是什么]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/112728826</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/112728826</guid><author>todingdong</author><pubDate>Sun, 17 Jan 2021 10:33:05 +0800</pubDate><description><![CDATA[文章目录李宏毅深度学习与NLP课程复旦神经网络课程唐宇迪：深度学习项目实战-Seq2Seq序列生模型+用RNN与LSTM网络原理进行唐诗生成1espresso阿婆主NLP从入门到放弃Transformer从零详细解读(可能是你见过最通俗易懂的讲解)BERT专题系列（二）：Transformer （Attention is all you need）

李宏毅深度学习与NLP课程
https://www.bilibili.com/video/BV1JE411g7XF?p=54
https://www.bili]]></description><category></category></item><item><title><![CDATA[综述：时间序列LB LM检验]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/112145770</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/112145770</guid><author>todingdong</author><pubDate>Sun, 03 Jan 2021 19:12:51 +0800</pubDate><description><![CDATA[序列自相关检验（也可以视为白噪声检验）
文献出处：
https://blog.nex3z.com/2019/07/24/time-series-q-statistic/
https://www.zhihu.com/question/22863169

作用
对于滞后相关的检验，我们常常采用的方法还包括计算ACF和PCAF并观察其图像，但是无论是ACF还是PACF都仅仅考虑是否存在某一特定滞后阶数的相关。LB检验则是基于一系列滞后阶数，判断序列总体的相关性或者说随机性是否存在。
ARIMA模型去拟合数据后我们]]></description><category></category></item><item><title><![CDATA[时间序列交叉验证]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/112093076</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/112093076</guid><author>todingdong</author><pubDate>Sat, 02 Jan 2021 04:04:50 +0800</pubDate><description><![CDATA[来源
https://www.mdpi.com/1099-4300/21/10/1015/htm#FD3-entropy-21-01015
翻译
#For cross-validation, we follow the time-series machine-learning literature and propose the use of rolling-origin evaluation [24], also known as rolling-origin-recalibration evaluati]]></description><category></category></item><item><title><![CDATA[树模型调参指南——官网文档]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/112001522</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/112001522</guid><author>todingdong</author><pubDate>Thu, 31 Dec 2020 02:51:29 +0800</pubDate><description><![CDATA[提示：文章写完后，目录可以自动生成，如何生成可参考右边的帮助文档

文章目录决策树随机森林GBDTXGboostLightGBMCatboost
决策树
随机森林
GBDT
XGboost
https://xgboost.readthedocs.io/en/latest/parameter.html
LightGBM
Catboost

...]]></description><category></category></item><item><title><![CDATA[关于时间序列の仓库]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/112001226</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/112001226</guid><author>todingdong</author><pubDate>Thu, 31 Dec 2020 01:02:09 +0800</pubDate><description><![CDATA[时间序列
Sun-Yize-SDUWH
https://github.com/Sun-Yize-SDUWH/Time-Series/blob/master/3_Time_Series_predict/jupyter%20notebook/sp100list_predict.ipynb
SP100list 时间序列数据预测
在以下的预测的python程序中，分别选取了时间序列与机器学习的相关算法，对数据分别进行预测，并将预测结果进行比对，选出较好的预测算法。
时间序列算法包括：AR模型，Theta模型，Dam]]></description><category></category></item><item><title><![CDATA[Pyhon里一些有趣的包——关于时间序列、回归、机器学习auto]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111999875</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111999875</guid><author>todingdong</author><pubDate>Wed, 30 Dec 2020 23:58:26 +0800</pubDate><description><![CDATA[文章目录==机器学习类==nyokasimple-learnautoml==时间序列新方法==fbprophet==时间序列自动化==pmdarimaAuto_TSarch
机器学习类
nyoka
https://softwareag.github.io/nyoka/
这个包似乎和scikit-learn有点像
simple-learn
自动ML
automl
自动ML
时间序列新方法
fbprophet
https://facebook.github.io/prophet/docs/quick_start]]></description><category></category></item><item><title><![CDATA[ESN学习笔记——echotorch(3)mackey_glass_esn]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111940414</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111940414</guid><author>todingdong</author><pubDate>Wed, 30 Dec 2020 01:01:57 +0800</pubDate><description><![CDATA[文章目录加载数据集参数Use CUDA?Mackey glass 数据集数据加载器ESN 细胞对每一批完成训练训练集数据在训练集上预测打印训练集误差测试集数据测试集上预测打印预测误差
加载
import torch
from echotorch.datasets.MackeyGlassDataset import MackeyGlassDataset
import echotorch.nn as etnn
import echotorch.utils
from torch.autograd import V]]></description><category></category></item><item><title><![CDATA[ESN学习笔记——echotorch(2)narma10]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111940322</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111940322</guid><author>todingdong</author><pubDate>Wed, 30 Dec 2020 00:54:15 +0800</pubDate><description><![CDATA[文章目录加载库训练样本长度测试样本长度有多少训练/测试样本批量大小(同时处理多少个样本?)超参数预测/目标长度使用 CUDA?人工种子初始化NARMA30 数据集数据加载器内部矩阵W生成输入权重Win生成Bias vector偏差向量生成创建一个Leaky-integrated ESN，用最小二乘训练算法，调用 esn = etrs.ESN(如果可能，在GPU中传输end if对每一批end for计算输出矩阵Wout来结束训练得到训练集中的第一个样本，转为Variable.用我们训练过的ESN做预测打印训]]></description><category></category></item><item><title><![CDATA[ESN学习笔记——echotorch(1)介绍]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111938660</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111938660</guid><author>todingdong</author><pubDate>Tue, 29 Dec 2020 22:40:47 +0800</pubDate><description><![CDATA[文章目录


]]></description><category></category></item><item><title><![CDATA[null]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111937624</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111937624</guid><author>todingdong</author><pubDate>Tue, 29 Dec 2020 21:38:32 +0800</pubDate><description><![CDATA[文章目录回归案例载入包训练集和测试集训练模型拟合预测预测结果和真实值可视化模型评价，测试集上均方误差
回归案例
载入包
# some_file.py
import sys
sys.path.insert(0, '../src/easyesn/')
import matplotlib.pyplot as plt
import numpy as np
from easyesn import RegressionESN
from easyesn import helper as hlp

训练集和测试集
设置输入]]></description><category></category></item><item><title><![CDATA[null]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111936877</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111936877</guid><author>todingdong</author><pubDate>Tue, 29 Dec 2020 20:49:31 +0800</pubDate><description><![CDATA[文章目录预测案例载入库设置训练集和测试集(Repeated) 一步预测训练预测把预测值和真实值一起可视化生成任务（略）
预测案例
载入库
import numpy as np
import sys
sys.path.insert(0, '../src/easyesn/')
from easyesn.optimizers import GradientOptimizer
from easyesn import PredictionESN
from easyesn.optimizers import GridS]]></description><category></category></item><item><title><![CDATA[ESN学习笔记——easyesn（1）包的介绍]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111936489</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111936489</guid><author>todingdong</author><pubDate>Tue, 29 Dec 2020 20:42:33 +0800</pubDate><description><![CDATA[文章目录easy esn注意教程载入训练预测用于自动超参数优化，导入一些包创建一个新对象要优化参数，创建一个优化器对象和它一起使用
easy esn
“easyesn"是一个使用回声状态网络（esn，也称为"储层计算”）的递归神经网络库，具有高度易用的api，紧密面向"sklearn"。它旨在通过提供基于梯度的超参数自动调整（岭回归惩罚、谱半径、泄漏率和反馈比例）以及瞬态时间估计的算法，尽可能容易地利用"esn"。此外，它还集成了使用时空"esn"来预测和分类几何扩展输入信号的能力。
“easyesn"可以]]></description><category></category></item><item><title><![CDATA[时间序列在RNN中的输入问题]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111936153</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111936153</guid><author>todingdong</author><pubDate>Tue, 29 Dec 2020 20:16:54 +0800</pubDate><description><![CDATA[文章目录方法一：历史值一起上方法二：主成分分析方法三：相关系数
方法一：历史值一起上
1维预测
参考PKU的Tensorflow2教程
Xt 作为 Y
【Xt-1，……Xt-10】过去10天的历史值作为X
这样有很多对【X,Y】来预测
多维预测
https://blog.youkuaiyun.com/weixin_39653948/article/details/105352010
参考这个，也是用历史值
【X1,X2,X3】t-1
【X1,X2,X3】t-2
【X1,X2,X3】t-3
这样作为输入X，yt作为输出
]]></description><category></category></item><item><title><![CDATA[ESN学习笔记——原理与超参数]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111935267</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111935267</guid><author>todingdong</author><pubDate>Tue, 29 Dec 2020 19:18:53 +0800</pubDate><description><![CDATA[文章目录参考文献地址ESN原理(略）超参数==储备池内部连接权谱半径ρ====储备池规模N====输入单元尺度因子IS。====储备池稀疏程度SD。====输出正则化因子==泄露积分型回声状态网络（Leaky ESN）==泄漏率α==
参考文献地址
ESN原理(略）
论坛里有不少文章，我就不搬了，看一下可以大致搞懂原理。
知网搜回声状态网络也有很多中文文献，不一定要去啃英文论文。
超参数
储备池内部连接权谱半径ρ
其为连接权矩阵W的绝对值最大的特征值，记为λmax，λmax&lt;1是保证网络稳定的必要条件]]></description><category></category></item><item><title><![CDATA[TCN 一维预测の笔记]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111878714</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111878714</guid><author>todingdong</author><pubDate>Tue, 29 Dec 2020 02:25:38 +0800</pubDate><description><![CDATA[文章目录来源数据代码加载包说明读取数据设定参数搭建模型设置优化器模型拟合可视化
来源
这个例子Support of tensorflow.keras instead of keras
https://github.com/philipperemy/keras-tcn/tree/master/tasks
数据
month	milk_production_pounds
1962-01	589
1962-02	561
1962-03	640
1962-04	656
1962-05	727
1962-06	697
]]></description><category></category></item><item><title><![CDATA[Keras-TCN的API笔记]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/111878636</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/111878636</guid><author>todingdong</author><pubDate>Tue, 29 Dec 2020 02:01:47 +0800</pubDate><description><![CDATA[文章目录APITCN优势API参数nb_filters: Integerkernel_sizedilationsnb_stackspaddinguse_skip_connectionsreturn_sequencesdropout_rateactivationkernel_initializeruse_batch_normkwargsInput shapeOutput shapeTCN支持的任务类型代码示例①载入模块②确定batch的大小，时间步，输入维度③得到训练集④搭建网络（没太懂这步，有待研究，备注瞎]]></description><category></category></item><item><title><![CDATA[R语言KNN资料]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/110496826</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/110496826</guid><author>todingdong</author><pubDate>Wed, 02 Dec 2020 19:48:48 +0800</pubDate><description><![CDATA[https://blog.youkuaiyun.com/LdyInG_/article/details/103513186
https://www.rdocumentation.org/packages/kknn/versions/1.3.1/topics/kknn
https://www.cnblogs.com/ljhdo/p/4531164.html
https://recomm.cnblogs.com/blogpost/5539723

]]></description><category></category></item><item><title><![CDATA[更改分类算法的阙值的资料]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/110494256</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/110494256</guid><author>todingdong</author><pubDate>Wed, 02 Dec 2020 18:26:44 +0800</pubDate><description><![CDATA[Python
https://stackoverflow.com/questions/35692059/how-to-change-threshold-for-precision-and-recall-in-python-scikit-learn
http://www.voidcn.com/article/p-vnrifweg-bvk.html

]]></description><category></category></item><item><title><![CDATA[日记——从云服务器开始神经网络 安装ananconda3]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/110300430</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/110300430</guid><author>todingdong</author><pubDate>Sun, 29 Nov 2020 03:17:02 +0800</pubDate><description><![CDATA[此乃2020.11月的尝试
mobaxterm安装
anaconda安装
关键是一般的云服务器，比如我这个学校的服务器安装的python是2.7的，已经很旧了
需要安装新的
我在网上找了一堆教程，报错报错报错！！！！！
最后正解是：自己下个anaconda.sh的文件
1清华镜像下载安装包：https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/
2 用mobaxterm传到某个文件夹
3 输入如下代码
cd /home/文件夹的云上路径
sh Anaco]]></description><category></category></item><item><title><![CDATA[优快云打印代码]]></title><link>https://blog.youkuaiyun.com/todingdong/article/details/110133824</link><guid>https://blog.youkuaiyun.com/todingdong/article/details/110133824</guid><author>todingdong</author><pubDate>Wed, 25 Nov 2020 16:09:14 +0800</pubDate><description><![CDATA[(function(){
    'use strict';
    var articleBox = $("div.article_content");
    articleBox.removeAttr("style");
    $("#btn-readmore").parent().remove();
    $("#side").remove();
    $("#comment_title, #comment_list, #comment_bar, #comment_form, .announc]]></description><category></category></item></channel></rss>