Just a reminder, 2 pm today, Plaza branch, see you then.
Brian will be presenting next time (tentatively scheduled for October
1st, 2pm) on ggplot2.
-Madelaine
All,
Below are my notes from our discussion today.
--Nathan
References
http://code.google.com/apis/predict/docs/hello_world.html
http://code.google.com/p/google-prediction-api-r-client/
Requirements
Google account
Prediction api
Storage api
Free - 100 req/day; 5 mb trained/day
Paid -10,000 req/month; $0.002/MB trained (100 MB maximum)
Billing enabled
Methods
Univariate response, classification or regression
Missing values - changed to blank or zero
Setup
R packages: json, RCurl, googlepredictionapi
GSutil and some simple bash scripts
Data Requirements
Example (i.e. target or response) in the first column
Features (i.e. predictors) in all other columns
CSV, characters quoted
No headers
R: write.table(iris[,c(5,1:4)],'iris.txt',col.names=F,row.names=F,sep=',')
Hello world
> gsutil mb gs://kcrug2
> gsutil cp iris.txt gs://kcrug
> ./oauth-train.sh kcrug/iris
> ./oauth-check-training.sh kcrug/iris
> ./oauth-predict.sh kcrug/iris "\"6,3,5\""
R (401 error)
require(rjson)
require(RCurl)
library(help=googlepredictionapi)
my.model <- PredictionApiTrain(data="gs://kcrug/iris",verbose=T)
R Wrappers (alternative)
library(help=predictionapirwrapper)
token <- GetAuthToken(email="nwste...@gmail.com", passwd="xxxxxx")
GoogleTrain(auth_token=token$Auth, mybucket="kcrug", mydata="iris")
GoogleTrainCheck(auth_token=token$Auth, mybucket="kcrug", mydata="iris")
GooglePredict(auth_token=token$Auth, mybucket="kcrug", mydata="iris", myinput=c(5,3,6))
On Sat, Aug 20, 2011 at 11:18 AM, arielpaulson <arielp...@gmail.com> wrote: