Decision #2
Decision #2
Conversation
|
Hi Hendrik, sorry for only getting around this now. I have only 2 questions regarding this PR: Also, you changed the Dynamic Bayesian Network quite a lot, instead of creating a new Dynamic Decision Network. Does the old DBN still work? Depending on the answer, I might just cherry pick parts of this PR and merge it, instead of the entire PR so that it is easier for me to merge the Decision Networks back into to root repository. |
|
Hey Jan, |
jpoeppel
left a comment
jpoeppel
left a comment
There was a problem hiding this comment.
Hey Hendrik,
thanks for the quick reply! When I check, the changed files, it shows some test functions to be removed. Maybe you would need to rebase first. But if that was not intentional, I can just take care of that when merging, so you would not need to change that, just wanted to make sure that there was no special reason behind that.
| class ContinousNode(unittest.TestCase): | ||
| pass | ||
|
|
||
| class UtilityNode(unittest.TestCase): |
There was a problem hiding this comment.
Github claims these lines here have been removed.
| self.assertTrue(v in f.values[p]) | ||
|
|
||
|
|
||
| def test_create_from_utility_node(self): |
There was a problem hiding this comment.
Here are also some that appear to be missing after the merge
|
Yeah it was not intentional. You can take care of it ! |
|
Great thanks for the quick reply! I will try to find the time to merge it (also into the base repository) within the next couple of weeks. |
…: testing max sum and adding transition probability
…ples. For every example to get the optimal decision both algorithms are now used (classical and max_sum)
…gaussian filter and the current believe state
… in calculating the inner product, when a decision Rule is given for ALL Decision Nodes. Transition Matrix can now also be a dictionary (better readability). Added the possibility to set the decision Nodes to the best action based on a skill believe of the parent node: This feature still has to be updated
…hm (For the next skill). Bugfixing: Deleted inter edges in the unrolled net after doing update. The Child of the inter edge Relationship had an exponentially growing CPD which lead to Memory Problems. Obviously we dont want the next node to have the inter_edges internalized in the unrolled net otherwise we cant make proper updates with the same Dimension we had in the beginning. The Unrolled function looks a bit ugly now i try to make it more compact in the future
2 participants
No description provided.