supports multinomial logistic 28softmax 29 and binomial logistic regression

1# supports multinomial logistic (softmax) and binomial logistic regression
2
3from pyspark.sql import Row
4from pyspark.ml.linearalg import Vectors
5bdf = sc.parallelize([
6  Row(label=1.0, weight=2.0, features=Vectors.dense(1.0)),
7  Row(label=0.0, weight=2.0, features=Vectors.sparse(1, [], []))]).toDF()
8blor = LogisticRegression(maxIter=5, regParam=0.01, weightCol="weight")
9blorModel = blor.fit(bdf)
10blorModel.coefficients
11# DenseVector([5.5...])
12blorModel.intercept
13# -2.68...
14mdf = sc.parallelize([
15  Row(label=1.0, weight=2.0, features=Vectors.dense(1.0)),
16  Row(label=0.0, weight=2.0, features=Vectors.sparse(1, [], [])),
17  Row(label=2.0, weight=2.0, features=Vectors.dense(3.0))]).toDF()
18mlor = LogisticRegression(maxIter=5, regParam=0.01, weightCol="weight",
19                          family="multinomial")
20mlorModel = mlor.fit(mdf)
21print(mlorModel.coefficientMatrix)
22# DenseMatrix([[-2.3...],
23#              [ 0.2...],
24#              [ 2.1... ]])
25mlorModel.interceptVector
26# DenseVector([2.0..., 0.8..., -2.8...])
27test0 = sc.parallelize([Row(features=Vectors.dense(-1.0))]).toDF()
28result = blorModel.transform(test0).head()
29result.prediction
30# 0.0
31result.probability
32# DenseVector([0.99..., 0.00...])
33result.rawPrediction
34# DenseVector([0.99..., 0.00...])
35test1 = sc.parallelize([Row(features=Vectors.sparse(1, [0], [1.0]))]).toDF()
36blorModel.transform(test1).head().prediction
371.0
38blor.setParams("vector")
39# Traceback (most recent call last):
40#     ...
41# TypeError: Method setParams forces keyword arguments.
42lr_path = temp_path + "/lr"
43blor.save(lr_path)
44lr2 = LogisticRegression.load(lr_path)
45lr2.getMaxIter()
46# 5
47model_path = temp_path + "/lr_path"
48blorModel.save(model_path)
49model2 = LogisticRegressionModel.load(model_path)
50blorModel.coefficients[0] == model2.coefficients[0]
51# True
52blorModel.intercept == model2.intercept
53# True