Github user mengxr commented on a diff in the pull request:
https://github.com/apache/spark/pull/79#discussion_r10957131
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/tree/impurity/Impurity.scala ---
@@ -0,0 +1,42 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.mllib.tree.impurity
+
+/**
+ * Trail for calculating information gain
+ */
+trait Impurity extends Serializable {
+
+ /**
+ * information calculation for binary classification
+ * @param c0 count of instances with label 0
+ * @param c1 count of instances with label 1
+ * @return information value
+ */
+ def calculate(c0 : Double, c1 : Double): Double
+
+ /**
+ * information calculation for regression
+ * @param count number of instances
+ * @param sum sum of labels
+ * @param sumSquares summation of squares of the labels
+ * @return information value
+ */
+ def calculate(count: Double, sum: Double, sumSquares: Double): Double
--- End diff --
The major loss of precision is from `sumSquares - sum*sum/count`, where a
large number subtracts another. Changing the interface to `(count, avg,
avgSquare)` would help avoid overflow but has nothing to do with precision. I
agree with @etrain that we can improve it in a future PR.
The question is whether we should make `calculate(c0, c1)` and
`calculate(count, sum, square)` a public method of `Impurity`. In either
classification or regression, `Impurity` works like an accumulator. What we
need is to describe how to process a label of type Double, how to merge two
`Impurity` instances, and how to get impurity from an instance, which is very
similar to `StatCounter`. It is strange to see Gini only implements the first
but not the second, while Variance only implements the second but not the
first. We probably need to reconsider the design here. For example, if we want
to handle three classes in the future, we will run into a signature collision
with `calculate(count, sum, squareSum)`.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
