[ML-4069] Improve timing of estimators (#161)

This gives the following running times:
```
recommendation.ALS	72.083s
classification.DecisionTreeClassification	37.125s
classification.DecisionTreeClassification	33.274s
regression.DecisionTreeRegression	31.252s
regression.DecisionTreeRegression	63.35s
fpm.FPGrowth	6.219s
fpm.FPGrowth	5.342s
classification.GBTClassification	46.154s
regression.GBTRegression	45.832s
clustering.GaussianMixture	18.936s
regression.GLMRegression	20.342s
clustering.KMeans	32.473s
clustering.LDA	44.574s
clustering.LDA	24.658s
classification.LinearSVC	39.84s
regression.LinearRegression	43.335s
classification.LogisticRegression	41.637s
classification.LogisticRegression	37.711s
classification.NaiveBayes	23.351s
classification.RandomForestClassification	20.781s
regression.RandomForestRegression	39.971s
feature.Word2Vec	51.892s
```

src/main/resources/com/databricks/spark/sql/perf/mllib/config/mllib-large.yaml

@@ -9,53 +9,58 @@ common:
 benchmarks:
   - name: classification.DecisionTreeClassification
     params:
-      depth: [5, 10]
+      depth: [5, 12]
       numClasses: 4
   - name: classification.GBTClassification
     params:
-      numFeatures: 3000
+      numFeatures: 1500
-      depth: 5
+      depth: 4
       numClasses: 4
-      maxIter: 10
+      maxIter: 8
   - name: classification.RandomForestClassification
     params:
       depth: 10
+      numFeatures: 1000
       numClasses: 4
-      maxIter: 200 # number of trees
+      maxIter: 100
   - name: classification.LogisticRegression
     params:
+      numExamples: 700000
+      numFeatures: 5000
+      elasticNetParam: [0.0, 0.5]
       regParam: 0.01
       tol: 0.0
-      maxIter: 20
+      maxIter: 10
   - name: classification.LinearSVC
     params:
+      numExamples: 500000
       regParam: 0.01
       tol: 0
-      maxIter: 20
+      maxIter: 10
   - name: classification.NaiveBayes
     params:
+      numExamples: 2000000
       numFeatures: 5000
-      numClasses: 2
+      numClasses: 10
       smoothing: 1.0
   - name: clustering.GaussianMixture
     params:
-      numExamples: 100000
+      numFeatures: 30
-      numTestExamples: 100000
+      k: 15
-      numFeatures: 1000
+      maxIter: 15
-      k: 10
+      tol: 0.0
-      maxIter: 10
-      tol: 0.01
   - name: clustering.KMeans
     params:
-      k: 50
+      k: 20
       maxIter: 20
       tol: 1e-3
   - name: clustering.LDA
     params:
+      numExamples: 200000
       docLength: 100
       vocabSize: 5000
-      k: 60
+      k: 20
-      maxIter: 20
+      maxIter: 10
       optimizer:
         - em
         - online
@@ -93,45 +98,48 @@ benchmarks:
     params:
       numExamples: 10000
       vocabSize: 100
-      docLength: 1000
+      docLength: 300
       numSynonymsToFind: 3
   - name: fpm.FPGrowth
     params:
+      numExamples: 10000000
       numItems: 10000
-      itemSetSize: [4, 10]
+      itemSetSize: [4, 14]
   - name: recommendation.ALS
     params:
-      numExamples: 50000000
+      numExamples: 20000000
       numTestExamples: 50000000
-      numUsers: 6000000
+      numUsers: 4000000
-      numItems: 6000000
+      numItems: 4000000
       regParam: 0.01
       rank: 10
-      maxIter: 10
+      maxIter: 8
   - name: regression.DecisionTreeRegression
     params:
-      depth: [5, 10]
+      depth: [4, 7]
   - name: regression.GBTRegression
     params:
-      numFeatures: 2000
+      numFeatures: 1000
-      depth: 5
+      depth: 4
-      maxIter: 5
+      maxIter: 8
   - name: regression.GLMRegression
     params:
-      numExamples: 500000
+      numExamples: 400000
       numTestExamples: 500000
-      numFeatures: 1000
+      numFeatures: 500
       link: log
       family: gaussian
       tol: 0.0
-      maxIter: 10
+      maxIter: 8
       regParam: 0.1
   - name: regression.LinearRegression
     params:
+      numFeatures: 5000
       regParam: 0.01
       tol: 0.0
-      maxIter: 20
+      maxIter: 9
   - name: regression.RandomForestRegression
     params:
-      depth: 10
+      depth: 7
-      maxIter: 4
+      numFeatures: 500
+      maxIter: 16