[SPARK-20258][Doc][SparkR] Fix SparkR logistic regression example in programming guide (did not converge)

[actuaryzhang]

What changes were proposed in this pull request?

SparkR logistic regression example did not converge in programming guide (for IRWLS). All estimates are essentially zero:

training2 <- read.df("data/mllib/sample_binary_classification_data.txt", source = "libsvm")
df_list2 <- randomSplit(training2, c(7,3), 2)
binomialDF <- df_list2[[1]]
binomialTestDF <- df_list2[[2]]
binomialGLM <- spark.glm(binomialDF, label ~ features, family = "binomial")

17/04/07 11:42:03 WARN WeightedLeastSquares: Cholesky solver failed due to singular covariance matrix. Retrying with Quasi-Newton solver.

> summary(binomialGLM)

Coefficients:
                 Estimate
(Intercept)    9.0255e+00
features_0     0.0000e+00
features_1     0.0000e+00
features_2     0.0000e+00
features_3     0.0000e+00
features_4     0.0000e+00
features_5     0.0000e+00
features_6     0.0000e+00
features_7     0.0000e+00

[felixcheung]

LGTM

[felixcheung]

is this an issue with binary_classification_data data?

[actuaryzhang]

just a bad example.

[actuaryzhang]

@felixcheung
Just noticed that the current example for logistic regression in the programming guide did not seem to be a good one. It did not converge using IRWLS, and Quasi-Newton yielded almost zero estimates for all coefficients.

[felixcheung]

actually, you might need to leave it as relative path, ie. not starting with /

[actuaryzhang]

Thanks! copy paste error. Corrected now.

[SparkQA]

Test build #75613 has finished for PR 17571 at commit 95b5383.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #75615 has finished for PR 17571 at commit f7e71ea.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[felixcheung]

right, tests don't run example anyway...

[felixcheung]

merged to master.