search

prnicolas / ScalaMl

Project, source code and data files for 1st edition "Scala for Machine Learning"
http://www.scalaformachinelearning.com
151 stars 86 forks source link

Current rev102 PDF Chap1 documented results do not match current test execution #9

Open cycle23 opened 8 years ago

cycle23 commented 8 years ago

The end of Chapter 1 indicates we may expect to see a chart showing 3 different learning rates, and also that the classifications for the first two data points will be a "1" and "0", with predicted values of >~0.5 and <~0.1 respectively.

However, execution of the tests (just before today's update that appears to have broken some things), results in two classifications of "0" for those same data points, and predicted values of ~0.46 and ~0.25.

I also notice the text indicates the nIters would be output, but this is also not the case. 

 *****  test#1 LogBinRegressionEval binomial logistic regression
LogBinRegressionEval weights: -0.027411339959812955,-0.5610971521322289,-1.084359693720546
input 0.008837209302325677,4.10914E7 normalized 0.06306471253904308,0.06147527493891754 class (0,0.46768060497106606)|
input 0.06949806949806936,3.618657E8 normalized 0.5171000761944574,0.6415991941274024 class (0,0.2663386306666643)
LogBinRegressionEval completed
cycle23 commented 8 years ago

Note, the chapter also indicated an EPS of 0.02, which is invalid in the current limits (<0.01 required). I also tried ETA of 0.0005, and got closer results.