注册 登录  
 加关注
   显示下一条  |  关闭
温馨提示!由于新浪微博认证机制调整,您的新浪微博帐号绑定已过期,请重新绑定!立即重新绑定新浪微博》  |  关闭

Bioinformatics home

 
 
 

日志

 
 

Cancer classification using Google prediction API  

2011-11-11 22:15:45|  分类: Bioinformatics |  标签: |举报 |字号 订阅

  下载LOFTER 我的照片书  |
Google's prediction API is now available to everyone with nice tutorial on how to get going :) so I decided to check its performance using gene expression values for 5 selected genes using online-feature-selection from the classic AML/ALL dataset (training/test).
The dataset needed to be formatted (training/test) and copied to Google storage to get the default scripts working. 
Coming to the results, well the training step:
./oauth-train.sh mlani/top5_tr.csv
gave a performance of about 3/4:
./oauth-check-training.sh mlani/top5_tr.csv
{
 "kind": "prediction#training",
 "id": "mlani/top5_tr.csv",
 "selfLink": "https://www.googleapis.com/prediction/v1.2/training/mlani/top5_tr.csv",
 "modelInfo": {
  "modelType": "classification",
  "classificationAccuracy": 0.75
 },
 "trainingStatus": "DONE"
}
not too bad, considering they are using some sort of statistical learning and sample is not the recommended 10x of feature set... 
Coming to testing part:
 ./oauth-predict.sh mlani/top5_tr.csv "1122,178,847,33,1018"
{
 "kind": "prediction#output",
 "id": "mlani/top5_tr.csv",
 "selfLink": "https://www.googleapis.com/prediction/v1.2/training/mlani/top5_tr.csv/predict",
 "outputLabel": "ALL",
 "outputMulti": [
  {
   "label": "ALL",
   "score": 0.899588
  },
  {
   "label": "AML",
   "score": 0.100412
  }
 ]
}
so seems like the first example went well, putting ALL as ALL with ~90% score :)
To get it over all the test examples, needed to write a simple PERL script:
perl check-pred-googleapi.pl top5_te.csv 
ALLALLALLALLALLALLALLALLALLALLALLALLALLALLAML1
ALLAML2
ALLALLALLAMLAMLAMLAMLAMLALL3
AMLALL4
AMLAMLALL5
AMLAMLAML
Got 29 out of 34(accuracy:0.852941176470588)
which I guess is pretty good considering black box like usage :)
Happy predicting !
  评论这张
 
阅读(695)| 评论(0)
推荐 转载

历史上的今天

在LOFTER的更多文章

评论

<#--最新日志,群博日志--> <#--推荐日志--> <#--引用记录--> <#--博主推荐--> <#--随机阅读--> <#--首页推荐--> <#--历史上的今天--> <#--被推荐日志--> <#--上一篇,下一篇--> <#-- 热度 --> <#-- 网易新闻广告 --> <#--右边模块结构--> <#--评论模块结构--> <#--引用模块结构--> <#--博主发起的投票-->
 
 
 
 
 
 
 
 
 
 
 
 
 
 

页脚

网易公司版权所有 ©1997-2017