Exploration of layers for tokenization and parsing.

.gitignore

@@ -1,5 +1,6 @@
 *#*
 *~
+.vscode
 *.ba?k
 *.pyc
 __pycache__

naive-nlu/tree_nlu/knowledge_base.py

@@ -4,8 +4,7 @@ import logging
 from .session.org_mode import global_session as session
 
 from .atoms import Atom
-from . import parsing
-from . import tokenization
+from . import layered_model
 from . import knowledge_evaluation
 from .modifiable_property import is_modifiable_property
 import random
@@ -15,21 +14,6 @@ def diff_knowledge(before, after):
     return jsondiff.diff(before, after)
 
 
-def randomized_weighted_list(elements):
-    # Randomized
-    randomized = list(elements)
-    random.shuffle(randomized)
-
-    # And return only once
-    already_returned = set()
-    for e in randomized:
-        if e in already_returned:
-            continue
-
-        yield e
-        already_returned.add(e)
-
-
 
 class KnowledgeBase(object):
     def __init__(self, knowledge={}, examples=[], trained=[]):
@@ -37,41 +21,9 @@ class KnowledgeBase(object):
         self.originals = []
         self.examples = copy.copy(examples)
         self.trained = copy.copy(trained)
-        self.structural_elements = set()
-        self.token_chains = {}
-        self.tokens = set()
-
-    def add_token_pair(self, precedent, consequent):
-        self.add_token(precedent)
-        self.add_token(consequent)
-
-        if precedent not in self.token_chains:
-            self.token_chains[precedent] = []
-        self.token_chains[precedent].append(consequent)
-
-    def add_token(self, token):
-        self.tokens.add(token)
-        if (not isinstance(token, Atom)) and (token not in self.structural_elements):
-            session().annotate('Found new structural element “{}”'.format(token))
-            self.structural_elements.add(token)
-
-    def expected_token_after_precedent(self, precedent=None):
-        if precedent not in self.token_chains:  # If there's no known precedent, just return all tokens
-            return randomized_weighted_list(self.tokens)
-
-        return randomized_weighted_list(self.token_chains[precedent])
-
-    def train_tokenizer(self, example):
-        with session().log('Training tokenizer'):
-            session().annotate("Example: {}".format(example))
-            tokens = tokenization.integrate_tokenization(self, example)
-
-            # Integrate knowledge of concept
-            for token in tokens:
-                if not token in self.knowledge:
-                    self.knowledge[token] = {}
-
+        self.layers = layered_model.BaseModel(self)
 
+    ## Parsing
     def train(self, examples):
         knowledge_before = copy.deepcopy(self.knowledge)
         with session().log('Train'):
@@ -86,11 +38,12 @@ class KnowledgeBase(object):
                         self.act_upon(result)
 
                 with session().log("language integration"):
-                    tokens, decomposition, inferred_tree = parsing.integrate_language(self, example)
-                    session().annotate("Tokens: {}".format(tokens))
-                    session().annotate("Inferred tree: {}".format(inferred_tree))
+                    for tokens, decomposition, inferred_tree in self.layers.integrate(self, example):
+                        session().annotate("Tokens: {}".format(tokens))
+                        session().annotate("Inferred tree: {}".format(inferred_tree))
 
                 with session().log("full information integration"):
+                    tokens = self.layers.tokenization.tokenize(example['text'], return_one=True)
                     result = knowledge_evaluation.integrate_information(self.knowledge, {
                         "elements": tokens,
                         "decomposition": decomposition,
@@ -105,7 +58,7 @@ class KnowledgeBase(object):
 
                 # Reduce values
                 with session().log("reprocessing"):
-                    self.trained = parsing.reprocess_language_knowledge(self, self.examples)
+                    self.layers.reprocess(self.examples)
 
             knowledge_after = copy.deepcopy(self.knowledge)
             knowledge_diff_getter = lambda: diff_knowledge(knowledge_before,
@@ -113,19 +66,6 @@ class KnowledgeBase(object):
 
             return knowledge_diff_getter
 
-    def tokenize(self, row, return_one=True):
-        row = row.lower()
-        with session().log("Tokenize: {}".format(row)):
-            options = list(tokenization.to_tokens(self, row))
-            session().log("Results:\n{}".format('\n'.join(map(str, options))))
-
-            if return_one:
-                chosen = tokenization.pick_one_tokenization(options, self)
-                session().log("Chosen: “{}”".format(chosen))
-                self.train_tokenizer({'text': row, 'tokens': chosen})
-                return chosen
-            return options
-
     def process(self, row):
         knowledge_before = copy.deepcopy(self.knowledge)
         with session().log("Process: {}".format(row)):

naive-nlu/tree_nlu/layered_model.py

@@ -0,0 +1,47 @@
+from .layers import tokenization_layer
+from .layers import parsing_layer
+
+
+def make_yield_pipe(layers, knowledge_base, example):
+    if len(layers) < 1:
+        yield example
+        return
+
+    input_generator = make_yield_pipe(layers[:-1], knowledge_base, example)
+    for input in input_generator:
+        print("-->", input)
+        for d in list(layers[-1].integrate(knowledge_base, input)):
+            yield d
+
+
+class BaseModel:
+    def __init__(self, knowledge_base):
+        self.tokenization = tokenization_layer.TokenizationLayer(knowledge_base)
+        self.parsing = parsing_layer.ParsingLayer()
+
+        self.layers = [
+            self.tokenization,
+            self.parsing,
+        ]
+
+    def reprocess(self, examples):
+        for example in examples:
+            self._reprocess_single(example)
+
+    def _reprocess_single(self, example):
+        return 
+        pattern_examples = []
+        for i, sample in enumerate(examples):
+            other = examples[:i] + examples[i + 1:]
+            match = get_matching(sample, other)
+            if len(match) > 0:
+                sample = (match, sample[1],)
+            pattern_examples.append(sample)
+
+        return pattern_examples
+
+    def integrate(self, knowledge_base, example):
+        yield from make_yield_pipe(self.layers, knowledge_base, example)    
+
+    def tokenize(self, row, return_one=True):
+        return self.tokenization.to_tokens(row)

naive-nlu/tree_nlu/layers/parsing.py

@@ -1,18 +1,14 @@
 #!/usr/bin/env python
 
-from . import knowledge_evaluation
-from . import tokenization
-
-from . import depth_meter
-from .session.org_mode import global_session as session
+from ..session.org_mode import global_session as session
 import re
 import copy
 
 from functools import reduce
 from typing import List, Dict
-from .modifiable_property import ModifiableProperty
-from . import parameters
-from .atoms import Atom, a, is_atom
+from ..modifiable_property import ModifiableProperty
+from .. import parameters
+from ..atoms import Atom, a, is_atom
 
 def make_template(knowledge_base, tokens, parsed):
     matcher = list(tokens)
@@ -83,8 +79,8 @@ def integrate_language(knowledge_base, example):
     text = example["text"].lower()
     parsed = example["parsed"]
 
+    tokens = example['tokens']
     resolved_parsed = copy.deepcopy(parsed)
-    tokens = list(tokenization.pick_one_tokenization(tokenization.to_tokens(knowledge_base, text), knowledge_base))
 
     while True:
         session().annotate("P: {}".format(resolved_parsed))
@@ -95,14 +91,14 @@ def integrate_language(knowledge_base, example):
 
         for position, atom in lower_levels:
             with session().log("Atom {}".format(atom)):
+                result = None
                 similars = get_similar_tree(knowledge_base, atom, tokens)
                 for similar in similars:
                     result = build_remix_matrix(knowledge_base, tokens, atom, similar)
                     if result is not None:
                         break
 
-                if result is None:
-                    raise Exception("No match found")
+                return
                 remix, (start_bounds, end_bounds) = result
 
                 after_remix = apply_remix(tokens[len(start_bounds):-len(end_bounds)], remix)
@@ -147,7 +143,7 @@ def integrate_language(knowledge_base, example):
     session().annotate("M: {}".format(matcher))
     session().annotate("R: {}".format(result))
     session().annotate("---")
-    return tokens, matcher, result
+    yield tokens, matcher, result
 
 
 def apply_remix(tokens, remix):
@@ -319,7 +315,7 @@ def get_similar_tree(knowledge_base, atom, tokens):
 
     sorted_possibilities = sorted(sorted_possibilities, key=lambda p: p[3] * 100 + p[4], reverse=True)
     if len(sorted_possibilities) < 1:
-        return None
+        return []
 
     for i, possibility in enumerate(sorted_possibilities):
         similar_matcher, similar_result, similar_result_resolved, _atom_score, _token_score = possibility
@@ -369,20 +365,6 @@ def get_matching(sample, other):
     return matching
 
 
-def reprocess_language_knowledge(knowledge_base, examples):
-    examples = knowledge_base.examples + examples
-
-    pattern_examples = []
-    for i, sample in enumerate(examples):
-        other = examples[:i] + examples[i + 1:]
-        match = get_matching(sample, other)
-        if len(match) > 0:
-            sample = (match, sample[1],)
-        pattern_examples.append(sample)
-
-    return pattern_examples
-
-
 def reverse_remix(tree_section, remix):
     result_section = []
     offset = 0

naive-nlu/tree_nlu/layers/parsing_layer.py

@@ -0,0 +1,11 @@
+from . import parsing
+
+class ParsingLayer:
+    def __init__(self):
+        pass
+
+    def integrate(self, knowledge_base, example):
+        yield from parsing.integrate_language(knowledge_base, example)
+
+    def train(self, knowledge_base, example):
+        assert False

naive-nlu/tree_nlu/layers/tokenization.py

@@ -1,5 +1,5 @@
-from .session.org_mode import global_session as session
-from .atoms import Atom, a, is_atom
+from ..session.org_mode import global_session as session
+from ..atoms import Atom, a, is_atom
 
 def lookahead_for_tokens_or_strucutral_elements(knowledge_base, remaining):
     for se in knowledge_base.structural_elements:

naive-nlu/tree_nlu/layers/tokenization_layer.py

@@ -0,0 +1,84 @@
+from ..session.org_mode import global_session as session
+from ..atoms import Atom
+from . import tokenization
+import random
+import copy
+
+def randomized_weighted_list(elements):
+    # Randomized
+    randomized = list(elements)
+    random.shuffle(randomized)
+
+    # And return only once
+    already_returned = set()
+    for e in randomized:
+        if e in already_returned:
+            continue
+
+        yield e
+        already_returned.add(e)
+
+class TokenizationLayer:
+    def __init__(self, knowledge_base):
+        self.structural_elements = set()
+        self.token_chains = {}
+        self.tokens = set()
+        self.knowledge_base = knowledge_base
+        self.knowledge = knowledge_base.knowledge
+
+    def integrate(self, knowledge_base, data):
+        assert knowledge_base is self.knowledge_base
+
+        print(data)
+        assert 'text' in data
+        with session().log("Tokenize: {}".format(data['text'])):
+            for tokens in tokenization.to_tokens(self, data['text']):
+                data_with_row = copy.copy(data)
+                data_with_row['tokens'] = tokens
+                print(data_with_row)
+                yield data_with_row
+
+
+    def tokenize(self, row, return_one=True):
+        row = row.lower()
+        with session().log("Tokenize: {}".format(row)):
+            options = list(tokenization.to_tokens(self, row))
+            session().log("Results:\n{}".format('\n'.join(map(str, options))))
+
+            if return_one:
+                chosen = tokenization.pick_one_tokenization(options, self)
+                session().log("Chosen: “{}”".format(chosen))
+                self.train({'text': row, 'tokens': chosen})
+                return chosen
+            return options
+
+    ## Tokenization
+    def add_token_pair(self, precedent, consequent):
+        self.add_token(precedent)
+        self.add_token(consequent)
+
+        if precedent not in self.token_chains:
+            self.token_chains[precedent] = []
+        self.token_chains[precedent].append(consequent)
+
+    def add_token(self, token):
+        self.tokens.add(token)
+        if (not isinstance(token, Atom)) and (token not in self.structural_elements):
+            session().annotate('Found new structural element “{}”'.format(token))
+            self.structural_elements.add(token)
+
+    def expected_token_after_precedent(self, precedent=None):
+        if precedent not in self.token_chains:  # If there's no known precedent, just return all tokens
+            return randomized_weighted_list(self.tokens)
+
+        return randomized_weighted_list(self.token_chains[precedent])
+
+    def train(self, example):
+        with session().log('Training tokenizer'):
+            session().annotate("Example: {}".format(example))
+            tokens = tokenization.integrate_tokenization(self, example)
+
+            # Integrate knowledge of concept
+            for token in tokens:
+                if not token in self.knowledge:
+                    self.knowledge[token] = {}

naive-nlu/tree_nlu/tests/gac_100.py

@@ -99,14 +99,14 @@ examples = [
              lambda knowledge: _assert('electricity' in knowledge.knowledge['computers']['performs-over']['use'])
          ),],
      }),
-    ('full_example',
-     {
-         "text": "The dominant language in france is french?",
-         "affirmation": "The dominant language in france is french",
-         "parsed": ("question",
-                    ("property-has-value", "france", "dominant-language", "french")),
-         "answer": True,
-     }),
+    # ('full_example',
+    #  {
+    #      "text": "The dominant language in france is french?",
+    #      "affirmation": "The dominant language in france is french",
+    #      "parsed": ("question",
+    #                 ("property-has-value", "france", "dominant-language", "french")),
+    #      "answer": True,
+    #  }),
     # {
     #     "text": "was abraham lincoln once president of the united states?",
     #     "affirmation": "was abraham lincoln once president of the united states?",

naive-nlu/tree_nlu/tests/tokenization.py

@@ -63,11 +63,11 @@ def main():
         show_progbar(i, total, example['text'])
         if case_type == 'example':
             with session().log(example['text']):
-                knowledge.train_tokenizer(example)
+                knowledge.layers.tokenization.train(example)
 
         elif case_type == 'test':
             with session().log(example['text']):
-                tokens = list(knowledge.tokenize(example['text']))
+                tokens = list(knowledge.layers.tokenization.tokenize(example['text']))
 
                 session().log('Expected “{}”, found “{}”'
                             .format(example['tokens'], tokens))

naive-nlu/tree_nlu/utils/tokenization.py

@@ -26,4 +26,4 @@ BASIC_TOKENIZATION_EXAMPLES = (
 def train_basic_tokenization(knowledge_base):
     with session().log('Training basic tokenization'):
         for example in BASIC_TOKENIZATION_EXAMPLES:
-            knowledge_base.train_tokenizer(example)
+            knowledge_base.layers.tokenization.train(example)