Hi Ted, Please update once SGD parsing issue is fixed.
Thanks Rajesh On Wed, Oct 17, 2012 at 2:22 PM, Rajesh Nikam <[email protected]> wrote: > Hello Ted, > > Thanks for investigating into it. > I would look forward for further analysis and fix in SGD. > > I appreciate your efforts in looking into it. > > Thanks, > Rajesh > > > > On Tue, Oct 16, 2012 at 10:23 PM, Ted Dunning <[email protected]>wrote: > >> Rajesh, >> >> In the testing that I did, I ran 100, 1000 and 10,000 passes through the >> data. All produced identical results. Thus it isn't an issue of SGD >> converging. >> >> I also did a parameter scan of lambda and saw no effect. >> >> I also did the standard thing in R with glm and got the expected (correct) >> results. >> >> I haven't looked yet in detail, but I really suspect that the reading of >> the data is horked. This is exactly how that behaves. >> >> On Tue, Oct 16, 2012 at 4:49 AM, Rajesh Nikam <[email protected]> >> wrote: >> >> > Hi Ted, >> > >> > I was thinking, this might be due to having only 100 instances for >> > training. >> > >> > So I have created test set with two classes having ~49K instances, >> included >> > all features as predictors. >> > PFA sgd.grps.zip with test file. >> > >> > mahout trainlogistic --input /usr/local/mahout/trainme/sgd-grps.csv >> > --output /usr/local/mahout/trainme/sgd-grps.model --target class >> > --categories 2 --features 128 --types n --predictors a1 a2 a3 a4 a5 a6 >> a7 >> > a8 a9 a10 a11 a12 a13 a14 a15 a16 a17 a18 a19 a20 a21 a22 a23 a24 a25 >> a26 >> > a27 a28 a29 a30 a31 a32 a33 a34 a35 a36 a37 a38 a39 a40 a41 a42 a43 a44 >> a45 >> > a46 a47 a48 a49 a50 a51 a52 a53 a54 a55 a56 a57 a58 a59 a60 a61 a62 a63 >> a64 >> > a65 a66 a67 a68 a69 a70 a71 a72 a73 a74 a75 a76 a77 a78 a79 a80 a81 a82 >> a83 >> > a84 a85 a86 a87 a88 a89 a90 a91 a92 a93 a94 a95 a96 a97 a98 a99 a100 >> a101 >> > a102 a103 a104 a105 a106 a107 a108 a109 a110 a111 a112 a113 a114 a115 >> a116 >> > a117 a118 a119 a120 a121 a122 a123 a124 a125 a126 a127 >> > >> > >> > mahout runlogistic --input /usr/local/mahout/trainme/sgd-grps.csv >> --model >> > /usr/local/mahout/trainme/sgd-grps.model --auc --confusion >> > >> > Still the results are similar, it classifies everything as class_1. >> > >> > AUC = 0.50 >> > confusion: [[*26563.0, 23006.0*], [0.0, 0.0]] >> > entropy: [[-0.0, -0.0], [-46.1, -21.4]] >> > >> > I am not sure why this is failing all the time. >> > >> > Looking forward for your reply. >> > >> > Thanks >> > Rajesh >> > >> > >> > >> > On Tue, Oct 16, 2012 at 3:57 AM, Ted Dunning <[email protected]> >> > wrote: >> > >> > > I would love to help and will before long. Just can't do it in the >> first >> > > part of this week. >> > > >> > > On Mon, Oct 15, 2012 at 6:28 AM, Rajesh Nikam <[email protected]> >> > > wrote: >> > > >> > > > Hello, >> > > > >> > > > I have asked below question on issue with using sgd on mahout forum. >> > > > >> > > > Similar issue with sgd is reported by >> > > > >> > > > >> > > >> > >> http://stackoverflow.com/questions/11221436/using-sgd-classifier-in-mahout >> > > > >> > > > Even below link has similar output: >> > > > >> > > > AUC = 0.57*confusion: [[27.0, 13.0], [0.0, 0.0]]* >> > > > entropy: [[-0.4, -0.3], [-1.2, -0.7]] >> > > > >> > > > >> > > > >> > http://sujitpal.blogspot.in/2012/09/learning-mahout-classification.html >> > > > >> > > > I am still wannder confusion how then this model works and used by >> > many ? >> > > > Not able to get any points on how to use SGD that generates >> effective >> > > > model. >> > > > >> > > > Could someone point out what is missing in input file or provided >> > > > parameters. >> > > > >> > > > I appreciate your help. >> > > > >> > > > Below is description of steps that I followed. >> > > > >> > > > PF Attached uses input files for experiment. >> > > > >> > > > I am using Iris Plants Database from Michael Marshall. PFA >> iris.arff. >> > > > Converted this to csv file just by updating header: >> iris-3-classes.csv >> > > > >> > > > mahout org.apache.mahout.classifier. >> > > > sgd.TrainLogistic --input >> > > /usr/local/mahout/trunk/*iris-3-classes.csv*--features 4 --output >> > > /usr/local/mahout/trunk/ >> > > > *iris-3-classes.model* --target class *--categories 3* --predictors >> > > > sepallength sepalwidth petallength petalwidth --types n >> > > > >> > > > >> it gave following error. >> > > > Exception in thread "main" java.lang.IllegalArgumentException: Can >> only >> > > > call classifyScalar with two categories >> > > > >> > > > Now created csv with only 2 classes. PFA iris-2-classes.csv >> > > > >> > > > >> trained iris-2-classes.csv with sgd >> > > > >> > > > mahout org.apache.mahout.classifier.sgd.TrainLogistic --input >> > > > /usr/local/mahout/trunk/*iris-2-classes.csv* --features 4 --output >> > > > /usr/local/mahout/trunk/*iris-2-classes.mode*l --target class >> > > *--categories >> > > > 2* --predictors sepallength sepalwidth petallength petalwidth >> --types n >> > > > >> > > > mahout runlogistic --input >> /usr/local/mahout/trunk/iris-2-classes.csv >> > > > --model /usr/local/mahout/trunk/iris-2-classes.model --auc >> --confusion >> > > > >> > > > AUC = 0.14 >> > > > confusion: [[50.0, 50.0], [0.0, 0.0]] >> > > > entropy: [[-0.6, -0.3], [-0.8, -0.4]] >> > > > >> > > > >> AUC seems to poor. Now changed --predictors >> > > > >> > > > mahout org.apache.mahout.classifier.sgd.TrainLogistic --input >> > > > /usr/local/mahout/trunk/*iris-2-classes.csv* --features 4 --output >> > > > /usr/local/mahout/trunk/*iris-2-classes.mode*l --target class >> > > *--categories >> > > > 2* --predictors sepalwidth petallength --types n >> > > > >> > > > mahout runlogistic --input >> /usr/local/mahout/trunk/iris-2-classes.csv >> > > > --model /usr/local/mahout/trunk/iris-2-classes.model --auc >> --confusion >> > > > --scores >> > > > >> > > > AUC = 0.80 >> > > > *confusion: [[50.0, 50.0], [0.0, 0.0]]* >> > > > entropy: [[-0.7, -0.3], [-0.7, -0.4]] >> > > > >> > > > This model classifies everything as category 1 which of no use. >> > > > >> > > > Thanks >> > > > Rajesh >> > > > >> > > > >> > > > >> > > > >> > > >> > >> > >
