<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[allen的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/fish2009122</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; fish2009122]]></copyright><item><title><![CDATA[本体在企业知识中台中的应用：技术原理与落地效果]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/163314627</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/163314627</guid><author>fish2009122</author><pubDate>Wed, 29 Jul 2026 23:36:36 +0800</pubDate><description><![CDATA[在知识工程里，本体是对某一领域中概念（类）、属性、关系以及约束的显式、形式化规范。类（Class）：如产品客户订单物料；属性（Property）：如产品价格客户所属行业；关系（Relation）：如属于供应上下游因果关系；实例（Instance）：具体的"iPhone 15"“某客户 A”；公理（Axiom）：如"一个订单必须关联一个客户""供应商不能是自身的客户"等约束规则。业界常用RDFS / OWL描述本体，用SPARQL进行语义查询，用Protege等工具进行可视化建模。]]></description><category></category></item><item><title><![CDATA[Moltbot(Clawdbot)]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/157515414</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/157515414</guid><author>fish2009122</author><pubDate>Thu, 29 Jan 2026 19:21:11 +0800</pubDate><description><![CDATA[Moltbot 安装]]></description><category></category></item><item><title><![CDATA[OpenWebUI by docker]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/146409169</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/146409169</guid><author>fish2009122</author><pubDate>Thu, 20 Mar 2025 23:50:36 +0800</pubDate><description><![CDATA[OpenWebUI。]]></description><category></category></item><item><title><![CDATA[docker启动报错：Cannot connect to the Docker daemon]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/132046600</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/132046600</guid><author>fish2009122</author><pubDate>Tue, 01 Aug 2023 17:13:49 +0800</pubDate><description><![CDATA[docker启动异常排查]]></description><category></category></item><item><title><![CDATA[DataHub Docker安装 &PostreSQL元数据集成]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/128290476</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/128290476</guid><author>fish2009122</author><pubDate>Mon, 12 Dec 2022 18:29:48 +0800</pubDate><description><![CDATA[install docker & docker-compose，包括docker-compose升级。读取 postgresql 元数据。]]></description><category></category></item><item><title><![CDATA[Jax, Numpy & Scipy]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/113826936</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/113826936</guid><author>fish2009122</author><pubDate>Tue, 16 Feb 2021 18:24:40 +0800</pubDate><description><![CDATA[Google开源的Jax据说很厉害，主要是用来解决numpy在GPU上支持度不太友好的开源包(PyTorch，Cupy，Numba等均同此)，今天小试一下，确实有优秀的地方
测试(仅限很少部分的numpy api，文档中有Scipy的api)


Version



随机数 &amp; jax替换numpy的点乘



jit




函数



梯度计算



Reference

官方文档
Github


...]]></description><category></category></item><item><title><![CDATA[SQL窗口分析函数]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/113110312</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/113110312</guid><author>fish2009122</author><pubDate>Mon, 25 Jan 2021 15:40:50 +0800</pubDate><description><![CDATA[Reference

Hive

Hive窗口函数
Hive Wiki WindowingAndAnalytics


Oracle

Window Function
Analytic Functions


MySQL

MySQL窗口函数
Window Functions



概述

代码准备(hive)

-- ALTER TABLE order RENAME TO order_infos
create table order_infos(name varchar(20), order_date d]]></description><category></category></item><item><title><![CDATA[docker-compose资源限制&docker资源监控]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/112004464</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/112004464</guid><author>fish2009122</author><pubDate>Thu, 31 Dec 2020 11:24:06 +0800</pubDate><description><![CDATA[资源限制设置

v2：pu_shares，cpu_quota，cpuset， mem_limit，memswap_limit，mem_swappines

docker-compose.yml  version: '2'
  services:
    testservice:
      image: image
      container_name: container_name
      ports:
        - 'ports_out:ports'
      privileged: T]]></description><category></category></item><item><title><![CDATA[模型评估指标]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/111866681</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/111866681</guid><author>fish2009122</author><pubDate>Mon, 28 Dec 2020 14:52:42 +0800</pubDate><description><![CDATA[分类：混淆矩阵/列联表

二分类：准确率，召回率，AUC（ROC曲线下的面积），logloss（对预测概率的似然估计），accuracy（概率阈值影响该指标），precision（概率阈值影响该指标）

TPTPTP：正预测为正(预测正确)
TNTNTN：负预测为负(预测正确)
FPFPFP：负预测为正
FNFNFN：正预测为负
预测结果为正的准确率：准确率/PPV=TPTP+FP\text{准确率/PPV} = \frac{TP}{TP + FP}准确率/PPV=TP+FPTP​
正样本预测的准确率：召回]]></description><category></category></item><item><title><![CDATA[假设检验小结]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/110138315</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/110138315</guid><author>fish2009122</author><pubDate>Wed, 25 Nov 2020 17:49:33 +0800</pubDate><description><![CDATA[Reference &amp; 相关，请点击此处


]]></description><category></category></item><item><title><![CDATA[统计功效(power)]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/110040002</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/110040002</guid><author>fish2009122</author><pubDate>Tue, 24 Nov 2020 11:33:57 +0800</pubDate><description><![CDATA[定义 &amp; 描述

别名 敏感度(sensitivity)
目的：反映一个检验检测出指定大小效应的能力
定义

在假设检验中， 拒绝原假设后， 接受正确的替换假设的概率
假设检验拒绝的漏报率：用同样的检验方法，对样本重采样/重复实验后计算检验值，统计检验值 &gt; 0.05(显著阈值)的占比
假设检验的“正确通过率”，1 - 漏报率
不犯第二类错误的概率
当AB两组差异真的存在时，我们能正确判断的概率


计算：1−β=Φ(z−z1−α/2)+Φ(−z−z1−α/2)1 - \beta = \Phi]]></description><category></category></item><item><title><![CDATA[keras layers笔记]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/108886065</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/108886065</guid><author>fish2009122</author><pubDate>Wed, 30 Sep 2020 16:24:11 +0800</pubDate><description><![CDATA[文章目录base`Layer.get_weights()`:Returns the current weights of the layer`Layer.set_weights(weights)`: Sets the weights of the layer, from Numpy arrays.`Model.get_config()`: config of the layer`Layer.add_loss(losses, **kwargs)`: Add loss tensor(s), potentiall]]></description><category></category></item><item><title><![CDATA[建模之常见激活函数（Keras，tf2）]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/108088120</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/108088120</guid><author>fish2009122</author><pubDate>Tue, 18 Aug 2020 22:36:41 +0800</pubDate><description><![CDATA[神经网络常见激活函数





序号
激活函数
表达式
优点
缺点
Keras
tf2




1
sigmoid
y=11+e−x=tanh(x/2)+12y=\frac{1}{1+e^{-x}}=\frac{\text{tanh}(x/2)+1}{2}y=1+e−x1​=2tanh(x/2)+1​
属于(0,1)(0, 1)(0,1)，即指定区间，又属于累计概率范围
指数计算成本不低；梯度弥散&amp;爆炸
keras.activations.sigmoid
tf.nn.sigmoid


2
elu.]]></description><category></category></item><item><title><![CDATA[建模之常见优化器（Keras）]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/108087465</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/108087465</guid><author>fish2009122</author><pubDate>Tue, 18 Aug 2020 21:21:01 +0800</pubDate><description><![CDATA[序号
优化函数
概述
场景
Keras




1
SGD
训练速度快容易陷于局部最优解

keras.optimizers.SGD：可设置动量和nesterov参数sgd


2
Adagrad(自适应学习率)
适用于数据稀疏或者分布不平衡的数据集中收敛速度快

keras.optimizers.Adagrad


3
RMSprop(自适应学习率)
修改了AdaGrad的梯度积累为指数加权的移动平均，使得其在非凸设定下效果更好。避免了学习率越来越低的的问题，而且能自适应地调节学习率
深度神经网络...]]></description><category></category></item><item><title><![CDATA[ClickHouse & OLAP]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/108057826</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/108057826</guid><author>fish2009122</author><pubDate>Mon, 17 Aug 2020 17:31:48 +0800</pubDate><description><![CDATA[install

docker install
connect by DBeaver

connect error: cannot create driver instance

原因：DBeaver下载ClickHouse驱动失败
解决方法：增加国内源的maven地址。窗口-&gt;首选项-&gt;DBeaver-&gt;驱动 -&gt; Maven。添加阿里云的maven地址http://maven.aliyun.com/nexus/content/groups/public/，并将其移到最上面，再次下]]></description><category></category></item><item><title><![CDATA[建模之常见模型评估指标（Keras, Sklearn, R）]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/108037833</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/108037833</guid><author>fish2009122</author><pubDate>Sun, 16 Aug 2020 22:24:43 +0800</pubDate><description><![CDATA[相同算法请见建模之常见损失函数




序号
概况
表达式
场景
keras
sklearn
R




1
分类准确率
预测正确个数的占比
分类
keras.metrics.Accuracyaccuracy
sklearn.metrics.accuracy_score
Accuracy


2
二分类准确率
同上
分类
keras.metrics.BinaryAccuracykeras.metrics.binary_accuracybinary_accuracy




3
多分类准确率
同上
分类
ke]]></description><category></category></item><item><title><![CDATA[建模之常见损失函数（Keras, Sklearn, R）]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/108020996</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/108020996</guid><author>fish2009122</author><pubDate>Sun, 16 Aug 2020 13:13:50 +0800</pubDate><description><![CDATA[序号
概况
表达式
场景
sklearn
keras
R




1
交叉熵
H(p,q)=∑x(p(x)×log(1q(x)))=∫xP(x)×log(Q(x))dxH(p, q)=\sum_x(p(x) \times \text{log}(\frac{1}{q(x)})) = \int_x P(x) \times \text{log}(Q(x))\mathrm{d}xH(p,q)=∑x​(p(x)×log(q(x)1​))=∫x​P(x)×log(Q(x))dx−log(p(y∣y^))=−(y...]]></description><category></category></item><item><title><![CDATA[PyTorch Documentation]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/107877842</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/107877842</guid><author>fish2009122</author><pubDate>Sat, 08 Aug 2020 12:17:58 +0800</pubDate><description><![CDATA[install packages
Documentation

官网
Github
PyTorch中文文档
ApacheCN




]]></description><category></category></item><item><title><![CDATA[傅里叶分析 & 应用]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/107727310</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/107727310</guid><author>fish2009122</author><pubDate>Sat, 01 Aug 2020 14:36:06 +0800</pubDate><description><![CDATA[傅里叶级数：Fourier Serie

傅里叶级数针对周期性函数：任意周期函数都可写成三角函数之和
公式

f(x)=a0×1+∑i=1∞(ai×cos(2×π×iT×x)+bi×sin(2×π×iT×x))=∑k=−∞∞(gk(x))f(x) = a_0  \times 1 + \sum_{i=1}^{\infty}(a_i \times cos(\frac{2 \times \pi \times i}{T} \times x) + b_i \times sin(\frac{2 \times \pi \]]></description><category></category></item><item><title><![CDATA[Sklearn之特征工程&建模]]></title><link>https://blog.youkuaiyun.com/fish2009122/article/details/107590577</link><guid>https://blog.youkuaiyun.com/fish2009122/article/details/107590577</guid><author>fish2009122</author><pubDate>Sun, 26 Jul 2020 11:03:19 +0800</pubDate><description><![CDATA[特征工程



sklearn建模



说明

特征工程来自于自己的总结，参考了使用sklearn做单机特征工程
sklearn来源于网络
以上如有侵权，请联系我




]]></description><category></category></item></channel></rss>