Traversing each example in the first generated regression tree is wrong

[Hugo101]

Take Dataset ToyCancer as an example. This is the first built regression tree (MaxDepth = 1):

   - [cancer(C):-  -0.16666666667,  cancer(C):- smokes(C) 0.5]

When building the second tree, we need the leaf node value for each example. As suggested in the ISSUE #11, 3 examples are split into left and right side respectively. Namely, three should be -1.666666, three should be 0.5.

However, This is the leaf node value for each example in the python version:

I have found where the specific problematic code is.
The problematic function is learnTree in tree.py

 node.learnedDecisionTree.sort(key = len)
 node.learnedDecisionTree = node.learnedDecisionTree[::-1] #reverse

Here sorting would not be correct because of so many decimal numbers which would be counted to the length of str. Based on the above code, the first tree would be:

  - [cancer(C):-  -0.16666666667,  cancer(C):- smokes(C) 0.5]

Therefore, all leaf node values for each example would be -0.166666666667.

Solution: keep the value of each clause has the same number of decimals.

Do some modification in function expandOnBestTest in tree.py.
the original code:

if clause[-1]!='-':
    node.learnedDecisionTree.append(clause[:-1]+" "+str(Utils.getleafValue(self.examples)))
else:
    node.learnedDecisionTree.append(clause+" "+str(Utils.getleafValue(self.examples)))

new modified code:

from decimal import getcontext, Decimal #added by Changbin
if clause[-1]!='-':
    leafValue = Decimal(Utils.getleafValue(self.examples)).quantize(Decimal('0.000000')) #added by Changbin
    node.learnedDecisionTree.append(clause[:-1]+" "+str(leafValue))          
 else:
    leafValue = Decimal(Utils.getleafValue(self.examples)).quantize(Decimal('0.000000')) #added by Changbin
    node.learnedDecisionTree.append(clause+" "+str(leafValue))          

new result is:

  - [cancer(C):- smokes(C) 0.500000,   cancer(C):-  -0.166667]

which could be used to search each example in the tree correctly.

[hayesall]

Resolved in fa9a104, closing issue.