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 > > > > > > > > > > > > > > >
