DLSS2016关于Andrew Ng报告
2017-03-06 16:37
148 查看
源github:https://github.com/thomasj02/DeepLearningProjectWorkflow
深度学习项目流程与工作细节
This document attempts to summarize Andrew Ng’s recommended machine learning workflow from his’s Nuts and Bolts of applying deeping learning “talk at deep learning Summer School 2016”.Any errors or misinterpretations are my own.
next 从这里开始
1measure human-level perfromance on your task.
2 do you training and test data come from the same distrbution?
yes or no ?
测量人的水平
The real goal of measuring humm-level perfromance is to est
测量人的水平表现
The real goal of measuring human-level performance is to estimate theBayes Error Rate. Knowing your Bayes Error Rate helps you figure out if your model is underfitting or overfitting your training data. More specifically, it will let us measure ‘Bias’ (as Ng defines it), which we use later in the workflow.
如果你的训练和测试数据来自于相同的分布
把你的数据整理和划分成Train / Dev / Test 集
Ng recommends a Train / Dev / Test split of approximately 70% / 15% / 15%.
衡量你的训练误差和开发设定误差,并计算拜厄斯和方差
Calculate your bias and variance as:
•Bias = (Training Set Error) - (Human Error)
•Variance = (Dev Set Error) - (Training Set Error)
出现High Bias了吗?首先修复它
An example of high bias:
Fix high biasbefore going on to the next step.
方差很高吗?修复它。
An example of high variance:
Once youFix Your High Variancethen you’re done!
如果你的训练和测试数据不是来自于相同的分布
划分你的数据
If your train and test data come from different distributions, make sure at least your dev and test sets are from the same distribution. You can do this by taking your test set and using half as dev and half as test.
Carve out a small portion of your training set (call thisTrain-Dev) and split your Test data intoDevandTest:
测量您的误差,并计算相关指标
Calculate these metrics to help know where to focus your efforts:
有高的Bias?修复它!
An example of high bias:
4.方差很高吗?修复它。
An example of high variance:
Fix your high variancebefore going on to the next step.
4.你的训练或测试不匹配吗?修复它
An example of train/test mismatch:
Fix Your Train/Test Mismatchbefore going on to the next step.
你的Dev Set出现过度拟合吗?修复它
An example of overfitting your dev set:
Once youfix your dev set overfitting, you’re done!
如何修复高Bias
Ng suggests these ways for fixing a model with high bias:
•Try a bigger model
•Try training longer
•Try a new model architecture (this can be hard)
如何修复高方差
Ng suggests these ways for fixing a model with high variance:
•Get more data
◾This includes data synthesis and data augmentation
•Try adding regularization
•Try early stopping
•Try new model architecture (this can be hard)
训练和测试失配,如何调整
Ng suggests these ways for fixing a model with high train/test mismatch:
•Try to get more data similar to your test data
•Try data synthesis and data augmentation
•Try new model architecture (this can be hard)
如何解决你Dev Set的过度拟合
Ng suggests only one way of fixing dev set overfitting:
•Get more dev data
Presumably this would include data synthesis and data augmentation as well.
深度学习项目流程与工作细节
This document attempts to summarize Andrew Ng’s recommended machine learning workflow from his’s Nuts and Bolts of applying deeping learning “talk at deep learning Summer School 2016”.Any errors or misinterpretations are my own.
next 从这里开始
1measure human-level perfromance on your task.
2 do you training and test data come from the same distrbution?
yes or no ?
测量人的水平
The real goal of measuring humm-level perfromance is to est
测量人的水平表现
The real goal of measuring human-level performance is to estimate theBayes Error Rate. Knowing your Bayes Error Rate helps you figure out if your model is underfitting or overfitting your training data. More specifically, it will let us measure ‘Bias’ (as Ng defines it), which we use later in the workflow.
如果你的训练和测试数据来自于相同的分布
把你的数据整理和划分成Train / Dev / Test 集
Ng recommends a Train / Dev / Test split of approximately 70% / 15% / 15%.
衡量你的训练误差和开发设定误差,并计算拜厄斯和方差
Calculate your bias and variance as:
•Bias = (Training Set Error) - (Human Error)
•Variance = (Dev Set Error) - (Training Set Error)
出现High Bias了吗?首先修复它
An example of high bias:
Fix high biasbefore going on to the next step.
方差很高吗?修复它。
An example of high variance:
Once youFix Your High Variancethen you’re done!
如果你的训练和测试数据不是来自于相同的分布
划分你的数据
If your train and test data come from different distributions, make sure at least your dev and test sets are from the same distribution. You can do this by taking your test set and using half as dev and half as test.
Carve out a small portion of your training set (call thisTrain-Dev) and split your Test data intoDevandTest:
测量您的误差,并计算相关指标
Calculate these metrics to help know where to focus your efforts:
有高的Bias?修复它!
An example of high bias:
4.方差很高吗?修复它。
An example of high variance:
Fix your high variancebefore going on to the next step.
4.你的训练或测试不匹配吗?修复它
An example of train/test mismatch:
Fix Your Train/Test Mismatchbefore going on to the next step.
你的Dev Set出现过度拟合吗?修复它
An example of overfitting your dev set:
Once youfix your dev set overfitting, you’re done!
如何修复高Bias
Ng suggests these ways for fixing a model with high bias:
•Try a bigger model
•Try training longer
•Try a new model architecture (this can be hard)
如何修复高方差
Ng suggests these ways for fixing a model with high variance:
•Get more data
◾This includes data synthesis and data augmentation
•Try adding regularization
•Try early stopping
•Try new model architecture (this can be hard)
训练和测试失配,如何调整
Ng suggests these ways for fixing a model with high train/test mismatch:
•Try to get more data similar to your test data
•Try data synthesis and data augmentation
•Try new model architecture (this can be hard)
如何解决你Dev Set的过度拟合
Ng suggests only one way of fixing dev set overfitting:
•Get more dev data
Presumably this would include data synthesis and data augmentation as well.
相关文章推荐
- 关于转正报告
- 关于SSL配置的报告
- 【转】 庞德致曹操:关于战胜关羽的可行性报告
- 关于X项目的报告
- 庞德致曹操:关于战胜关羽的可行性报告
- 关于康信邮箱运行情况报告。
- 关于现代可编程器件的读书报告(初稿)
- 关于俺是猪的分析报告:)
- 关于项目报告的写法问题
- 关于SSL配置的报告
- 关于Map和List的性能测试报告
- boost 智能指针 --- 关于性能的少数派报告
- 关于Kjava手机平台移植可行性报告
- 关于“上海.NET俱乐部”第一次技术交流会进展报告
- 关于SSL配置的报告
- 关于发表文章报告“必须填写文章正文”错误的问题
- 今天老大要我们写一个关于个人绩效的报告,呵呵,觉的下面老兄的写的不错。
- "关于2007权限问题的测试报告"的测试报告
- 20070812关于pdf报告
- 关于“三国众谋士”之IT从业可行性报告