Category Archives: Machine Learning

Phil 11.20.15

7:00 – Leave

Search: Optimal, Branch and Bound, A*
- Dead Horse principle.
- Admissible vs. consistency heuristics.
Dictionaries
- Wire up Add – Done. Also found a bug in the PHP code that was screwing up parent return. So that took a while…
- Wire up Delete – Done. That went much better!
- Wire up New Dictionary
- Wire up Extract
- Wire up Attach

7:00 – 5:00 Leave

Reasoning: Goal Trees and Rule-Based Expert Systems
- There, now I’m back in order.
- H. Simon – The complexity of the behavior is max(cplx(prgm), cplx(env))
- More Genesis – Elaboration graphs
- Genesis judges similarity in multiple ways: (this presentation, page 25)
  - Using word vectors
  - Using concept vectors: seeing similarities not evident in the words.
- Genesis aligns similar stories for analogical reasoning (Needleman-Wunch algorithm, which is a way of comparing string similarity using matrices)
IRB renewal – ask Wayne
In a fit of orderliness, created shortcuts to the cmd windows that the makefiles run in
Dictionaries
- Don’t forget to move the text-extraction calls to the methods that need it and see if that speeds up the others. Done. Much faster. PHP is dumb. Or needs a preloader/compiler
- Cleaned up the loading and display code a bit. Need to add a tree view (which looks like it can be done in pure CSS), but that can wait.
- Adding manual entry
  - Finished the form
  - Clear for entry and separate parent is done
  - addition
  - deletion
  - modification (adding parents in particular)
- Adding text extraction

7:00 – 5:00 Leave

Search: Depth-First, Hill Climbing, Beam
- Looks like I missed one. Will go back tomorrow.
- Patrick Winston is the instructor, and he showed a version of Genesis (around 42 minutes in), which maps stories into graphs and then searches through them at different levels of abstraction. It is capable of providing relevant answers to questions. Here’s a white paper. What I think is interesting here is the following:
  - How they describe their hierarchy (revenge, tit-for-tat, etc). This is built, not computed based on what looks to be a rules engine.
  - How text in a story can be looked at as a graph.
  - And that’s without reading the paper!
Working on dictionary directive
- Ramifying the dictionary choice through the controller. It should be set initially when the network loads (done), but should be changed if a new one is selected.
- There does need to be a second step where the DictPull is run and the text of the items is compared against the selected dictionary and KEYWORDS items are added to the network. This is when the network’s dictionary index should be changed.
- Downloading the dictionary.
- Need to upload a new word
12:00 workshop on Grant Writing in ITE 459
4:00 meeting with Dr. Pan.

7:00 – 4:00 leave

Reasoning: Goal Trees and Problem Solving (why is this pertinent to me?)
- Apply all safe transforms
- Apply heuristics
- AND nodes and OR nodes (AND/OR, problem reduction, or goal tree)
Realized that I have a redundant user_id in tn_dictionary_entries. This could be used to allow non-owners to add words to the dictionary. Which means that dictionaries could be shared. On the whole, I think that’s a good idea. Adding the change to the dictionary code.
Ok, back to text extraction (Is this a safe transform?)
Added a _buildExtractor() private method and imported all the extractor parts.
It does take a long time. Just to load? I’d like to try profiling…
Wow. Extracting, loading into the database, getting the JSON output and deleting the dictionary all work!
I think the next step is to either get some definitions or start building the directive. After my lunchtime ride, I decided that the directive is probably the best thing to do next. Adding user functionality is a good way of ensuring that the server functionality makes sense.
Added ‘get_user_dictionaries‘ to rssPull.php. We’ll start with that.
Got the skeleton of a directive up and retrieving the dictionary list.