Classifier objects will be removed in 5.0

A couple things I missed in 7f68dc6fc6

README.md

@@ -18,18 +18,19 @@ This will prompt for a string and classify it, then print (in JSON) a dict of
 the format `{category: probability, category:probability, ...}` to stdout. (For
 information about `-n <max_ngram_length>`, see section "Ngrams.")
 
-Alternatively, if you only need the most likely category, you can use this:
+### Checking individual words or ngrams
 
-    gptc classify [-n <max_ngram_length>] <-c|--category> <compiled model file>
+    gptc check <compiled model file> <token or ngram>
 
-This will prompt for a string and classify it, outputting the category on
-stdout (or "None" if it cannot determine anything).
+This is very similar to `gptc classify`, except it takes the input as an
+argument, and it treats the input as a single token or ngram.
 
 ### Compiling models
 
-    gptc compile [-n <max_ngram_length>] [-c <min_count>] <raw model file>
+    gptc compile [-n <max_ngram_length>] [-c <min_count>] <raw model file> <compiled model file>
 
-This will print the compiled model encoded in binary format to stdout.
+This will write the compiled model encoded in binary format to `<compiled model
+file>`.
 
 If `-c` is specified, words and ngrams used less than `min_count` times will be
 excluded from the compiled model.
@@ -43,14 +44,15 @@ example of the format. Any exceptions will be printed to stderr.
 
 ## Library
 
-### `gptc.Classifier(model, max_ngram_length=1)`
+### `Model.serialize(file)`
 
-Create a `Classifier` object using the given compiled model (as a `gptc.Model`
-object, not as a serialized byte string).
+Write binary data representing the model to `file`.
 
-For information about `max_ngram_length`, see section "Ngrams."
+### `Model.deserialize(encoded_model)`
 
-#### `Classifier.confidence(text)`
+Deserialize a `Model` from a file containing data from `Model.serialize()`.
+
+### `Model.confidence(text, max_ngram_length)`
 
 Classify `text`. Returns a dict of the format `{category: probability,
 category:probability, ...}`
@@ -60,16 +62,15 @@ common words between the input and the training data (likely, for example, with
 input in a different language from the training data), an empty dict will be
 returned.
 
-#### `Classifier.classify(text)`
+For information about `max_ngram_length`, see section "Ngrams."
 
-Classify `text`. Returns the category into which the text is placed (as a
-string), or `None` when it cannot classify the text.
+### `Model.get(token)`
 
-#### `Classifier.model`
+Return a confidence dict for the given token or ngram. This function is very
+similar to `Model.confidence()`, except it treats the input as a single token
+or ngram.
 
-The classifier's model.
-
-### `gptc.compile(raw_model, max_ngram_length=1, min_count=1)`
+### `Model.compile(raw_model, max_ngram_length=1, min_count=1, hash_algorithm="sha256")`
 
 Compile a raw model (as a list, not JSON) and return the compiled model (as a
 `gptc.Model` object).
@@ -79,15 +80,27 @@ For information about `max_ngram_length`, see section "Ngrams."
 Words or ngrams used less than `min_count` times throughout the input text are
 excluded from the model.
 
-### `gptc.Model.serialize()`
+The hash algorithm should be left as the default, which may change with a minor
+version update, but it can be changed by the application if needed. It is
+stored in the model, so changing the algorithm does not affect compatibility.
+The following algorithms are supported:
 
-Returns a `bytes` representing the model.
+* `md5`
+* `sha1`
+* `sha224`
+* `sha256`
+* `sha384`
+* `sha512`
+* `sha3_224`
+* `sha3_384`
+* `sha3_256`
+* `sha3_512`
+* `shake_128`
+* `shake_256`
+* `blake2b`
+* `blake2s`
 
-### `gptc.deserialize(encoded_model)`
-
-Deserialize a `Model` from a `bytes` returned by `Model.serialize()`.
-
-### `gptc.pack(directory, print_exceptions=False)
+### `gptc.pack(directory, print_exceptions=False)`
 
 Pack the model in `directory` and return a tuple of the format:
 
@@ -99,6 +112,13 @@ GPTC.
 
 See `models/unpacked/` for an example of the format.
 
+### `gptc.Classifier(model, max_ngram_length=1)`
+
+`Classifier` objects are deprecated starting with GPTC 3.1.0, and will be
+removed in 5.0.0. See [the README from
+3.0.2](https://git.kj7rrv.com/kj7rrv/gptc/src/tag/v3.0.1/README.md) if you need
+documentation.
+
 ## Ngrams
 
 GPTC optionally supports using ngrams to improve classification accuracy. They
@@ -118,7 +138,8 @@ reduced to the one used when compiling the model.
 
 ## Model format
 
-This section explains the raw model format, which is how models are created and edited.
+This section explains the raw model format, which is how models are created and
+edited.
 
 Raw models are formatted as a list of dicts. See below for the format:
 
@@ -129,9 +150,10 @@ Raw models are formatted as a list of dicts. See below for the format:
         }
     ]
 
-GPTC handles raw models as `list`s of `dict`s of `str`s (`List[Dict[str, str]]`), and they can be stored
-in any way these Python objects can be. However, it is recommended to store
-them in JSON format for compatibility with the command-line tool.
+GPTC handles raw models as `list`s of `dict`s of `str`s (`List[Dict[str,
+str]]`), and they can be stored in any way these Python objects can be.
+However, it is recommended to store them in JSON format for compatibility with
+the command-line tool.
 
 ## Emoji
 

benchmark.py

@@ -25,7 +25,7 @@ print(
     round(
         1000000
         * timeit.timeit(
-            "gptc.compile(raw_model, max_ngram_length)",
+            "gptc.Model.compile(raw_model, max_ngram_length)",
             number=compile_iterations,
             globals=globals(),
         )

gptc/__init__.py

@@ -2,13 +2,11 @@
 
 """General-Purpose Text Classifier"""
 
-from gptc.compiler import compile as compile
-from gptc.classifier import Classifier as Classifier
-from gptc.pack import pack as pack
-from gptc.tokenizer import has_emoji as has_emoji
-from gptc.model import Model as Model, deserialize as deserialize
+from gptc.pack import pack
+from gptc.model import Model
+from gptc.tokenizer import normalize
 from gptc.exceptions import (
-    GPTCError as GPTCError,
-    ModelError as ModelError,
-    InvalidModelError as InvalidModelError,
+    GPTCError,
+    ModelError,
+    InvalidModelError,
 )

gptc/__main__.py

@@ -17,6 +17,9 @@ def main() -> None:
         "compile", help="compile a raw model"
     )
     compile_parser.add_argument("model", help="raw model to compile")
+    compile_parser.add_argument(
+        "out", help="name of file to write compiled model to"
+    )
     compile_parser.add_argument(
         "--max-ngram-length",
         "-n",
@@ -41,19 +44,12 @@ def main() -> None:
         type=int,
         default=1,
     )
-    group = classify_parser.add_mutually_exclusive_group()
-    group.add_argument(
-        "-j",
-        "--json",
-        help="output confidence dict as JSON (default)",
-        action="store_true",
-    )
-    group.add_argument(
-        "-c",
-        "--category",
-        help="output most likely category or `None`",
-        action="store_true",
+
+    check_parser = subparsers.add_parser(
+        "check", help="check one word or ngram in model"
     )
+    check_parser.add_argument("model", help="compiled model to use")
+    check_parser.add_argument("token", help="token or ngram to check")
 
     pack_parser = subparsers.add_parser(
         "pack", help="pack a model from a directory"
@@ -63,29 +59,27 @@ def main() -> None:
     args = parser.parse_args()
 
     if args.subparser_name == "compile":
-        with open(args.model, "r") as f:
-            model = json.load(f)
+        with open(args.model, "r", encoding="utf-8") as input_file:
+            model = json.load(input_file)
 
-        sys.stdout.buffer.write(
-            gptc.compile(
+        with open(args.out, "wb+") as output_file:
+            gptc.Model.compile(
                 model, args.max_ngram_length, args.min_count
-            ).serialize()
-        )
+            ).serialize(output_file)
     elif args.subparser_name == "classify":
-        with open(args.model, "rb") as f:
-            model = gptc.deserialize(f.read())
-
-        classifier = gptc.Classifier(model, args.max_ngram_length)
+        with open(args.model, "rb") as model_file:
+            model = gptc.Model.deserialize(model_file)
 
         if sys.stdin.isatty():
             text = input("Text to analyse: ")
         else:
             text = sys.stdin.read()
 
-        if args.category:
-            print(classifier.classify(text))
-        else:
-            print(json.dumps(classifier.confidence(text)))
+        print(json.dumps(model.confidence(text, args.max_ngram_length)))
+    elif args.subparser_name == "check":
+        with open(args.model, "rb") as model_file:
+            model = gptc.Model.deserialize(model_file)
+        print(json.dumps(model.get(args.token)))
     else:
         print(json.dumps(gptc.pack(args.model, True)[0]))
 

gptc/classifier.py

@@ -1,94 +0,0 @@
-# SPDX-License-Identifier: GPL-3.0-or-later
-
-import gptc.tokenizer, gptc.compiler, gptc.exceptions, gptc.weighting
-import warnings
-from typing import Dict, Union, cast, List
-
-
-class Classifier:
-    """A text classifier.
-
-    Parameters
-    ----------
-    model : dict
-        A compiled GPTC model.
-
-    max_ngram_length : int
-        The maximum ngram length to use when tokenizing input. If this is
-        greater than the value used when the model was compiled, it will be
-        silently lowered to that value.
-
-    Attributes
-    ----------
-    model : dict
-        The model used.
-
-    """
-
-    def __init__(self, model: gptc.model.Model, max_ngram_length: int = 1):
-        self.model = model
-        model_ngrams = model.max_ngram_length
-        self.max_ngram_length = min(max_ngram_length, model_ngrams)
-        self.has_emoji = gptc.tokenizer.has_emoji and model.has_emoji
-
-    def confidence(self, text: str) -> Dict[str, float]:
-        """Classify text with confidence.
-
-        Parameters
-        ----------
-        text : str
-            The text to classify
-
-        Returns
-        -------
-        dict
-            {category:probability, category:probability...} or {} if no words
-            matching any categories in the model were found
-
-        """
-
-        model = self.model.weights
-
-        tokens = gptc.tokenizer.tokenize(
-            text, self.max_ngram_length, self.has_emoji
-        )
-        numbered_probs: Dict[int, float] = {}
-        for word in tokens:
-            try:
-                weighted_numbers = gptc.weighting.weight(
-                    [i / 65535 for i in cast(List[float], model[word])]
-                )
-                for category, value in enumerate(weighted_numbers):
-                    try:
-                        numbered_probs[category] += value
-                    except KeyError:
-                        numbered_probs[category] = value
-            except KeyError:
-                pass
-        total = sum(numbered_probs.values())
-        probs: Dict[str, float] = {
-            self.model.names[category]: value / total
-            for category, value in numbered_probs.items()
-        }
-        return probs
-
-    def classify(self, text: str) -> Union[str, None]:
-        """Classify text.
-
-        Parameters
-        ----------
-        text : str
-            The text to classify
-
-        Returns
-        -------
-        str or None
-            The most likely category, or None if no words matching any
-            category in the model were found.
-
-        """
-        probs: Dict[str, float] = self.confidence(text)
-        try:
-            return sorted(probs.items(), key=lambda x: x[1])[-1][0]
-        except IndexError:
-            return None

gptc/compiler.py

@@ -1,86 +0,0 @@
-# SPDX-License-Identifier: GPL-3.0-or-later
-
-import gptc.tokenizer
-import gptc.model
-from typing import Iterable, Mapping, List, Dict, Union
-
-
-def compile(
-    raw_model: Iterable[Mapping[str, str]],
-    max_ngram_length: int = 1,
-    min_count: int = 1,
-) -> gptc.model.Model:
-    """Compile a raw model.
-
-    Parameters
-    ----------
-    raw_model : list of dict
-        A raw GPTC model.
-
-    max_ngram_length : int
-        Maximum ngram lenght to compile with.
-
-    Returns
-    -------
-    dict
-        A compiled GPTC model.
-
-    """
-
-    categories: Dict[str, List[int]] = {}
-
-    for portion in raw_model:
-        text = gptc.tokenizer.tokenize(portion["text"], max_ngram_length)
-        category = portion["category"]
-        try:
-            categories[category] += text
-        except KeyError:
-            categories[category] = text
-
-    word_counts: Dict[int, Dict[str, int]] = {}
-
-    names = []
-
-    for category, text in categories.items():
-        if not category in names:
-            names.append(category)
-
-        for word in text:
-            try:
-                counts_for_word = word_counts[word]
-            except KeyError:
-                counts_for_word = {}
-                word_counts[word] = counts_for_word
-
-            try:
-                word_counts[word][category] += 1
-            except KeyError:
-                word_counts[word][category] = 1
-
-    word_counts = {
-        word: counts
-        for word, counts in word_counts.items()
-        if sum(counts.values()) >= min_count
-    }
-
-    word_weights: Dict[int, Dict[str, float]] = {}
-    for word, values in word_counts.items():
-        for category, value in values.items():
-            try:
-                word_weights[word][category] = value / len(categories[category])
-            except KeyError:
-                word_weights[word] = {
-                    category: value / len(categories[category])
-                }
-
-    model: Dict[int, List[int]] = {}
-    for word, weights in word_weights.items():
-        total = sum(weights.values())
-        new_weights: List[int] = []
-        for category in names:
-            new_weights.append(
-                round((weights.get(category, 0) / total) * 65535)
-            )
-        model[word] = new_weights
-
-    return gptc.model.Model(model, names, max_ngram_length)

gptc/model.py

@@ -1,9 +1,120 @@
 # SPDX-License-Identifier: GPL-3.0-or-later
 
+from typing import (
+        Iterable,
+        Mapping,
+    List,
+    Dict,
+    cast,
+    BinaryIO,
+    Tuple,
+    TypedDict,
+)
+import json
 import gptc.tokenizer
 from gptc.exceptions import InvalidModelError
-from typing import Iterable, Mapping, List, Dict, Union
-import json
+import gptc.weighting
+
+def _count_words(
+    raw_model: Iterable[Mapping[str, str]],
+    max_ngram_length: int,
+    hash_algorithm: str,
+) -> Tuple[Dict[int, Dict[str, int]], Dict[str, int], List[str]]:
+    word_counts: Dict[int, Dict[str, int]] = {}
+    category_lengths: Dict[str, int] = {}
+    names: List[str] = []
+
+    for portion in raw_model:
+        text = gptc.tokenizer.hash_list(
+            gptc.tokenizer.tokenize(portion["text"], max_ngram_length),
+            hash_algorithm,
+        )
+        category = portion["category"]
+
+        if not category in names:
+            names.append(category)
+
+        category_lengths[category] = category_lengths.get(category, 0) + len(
+            text
+        )
+
+        for word in text:
+            if word in word_counts:
+                try:
+                    word_counts[word][category] += 1
+                except KeyError:
+                    word_counts[word][category] = 1
+            else:
+                word_counts[word] = {category: 1}
+
+    return word_counts, category_lengths, names
+
+
+def _get_weights(
+    min_count: int,
+    word_counts: Dict[int, Dict[str, int]],
+    category_lengths: Dict[str, int],
+    names: List[str],
+) -> Dict[int, List[int]]:
+    model: Dict[int, List[int]] = {}
+    for word, counts in word_counts.items():
+        if sum(counts.values()) >= min_count:
+            weights = {
+                category: value / category_lengths[category]
+                for category, value in counts.items()
+            }
+            total = sum(weights.values())
+            new_weights: List[int] = []
+            for category in names:
+                new_weights.append(
+                    round((weights.get(category, 0) / total) * 65535)
+                )
+            model[word] = new_weights
+    return model
+
+class ExplanationEntry(TypedDict):
+    weight: float
+    probabilities: Dict[str, float]
+    count: int
+
+
+Explanation = Dict[
+    str,
+    ExplanationEntry,
+]
+
+Log = List[Tuple[str, float, List[float]]]
+
+
+class Confidences(dict[str, float]):
+    def __init__(self, probs: Dict[str, float]):
+        dict.__init__(self, probs)
+
+
+class TransparentConfidences(Confidences):
+    def __init__(
+        self,
+        probs: Dict[str, float],
+        explanation: Explanation,
+    ):
+        self.explanation = explanation
+        Confidences.__init__(self, probs)
+
+
+def convert_log(log: Log, names: List[str]) -> Explanation:
+    explanation: Explanation = {}
+    for word2, weight, word_probs in log:
+        if word2 in explanation:
+            explanation[word2]["count"] += 1
+        else:
+            explanation[word2] = {
+                "weight": weight,
+                "probabilities": {
+                    name: word_probs[index] for index, name in enumerate(names)
+                },
+                "count": 1,
+            }
+    return explanation
 
 
 class Model:
@@ -12,78 +123,200 @@ class Model:
         weights: Dict[int, List[int]],
         names: List[str],
         max_ngram_length: int,
-        has_emoji: Union[None, bool] = None,
+        hash_algorithm: str,
     ):
         self.weights = weights
         self.names = names
         self.max_ngram_length = max_ngram_length
-        self.has_emoji = (
-            gptc.tokenizer.has_emoji if has_emoji is None else has_emoji
+        self.hash_algorithm = hash_algorithm
+
+    def confidence(
+        self, text: str, max_ngram_length: int, transparent: bool = False
+    ) -> Confidences:
+        """Classify text with confidence.
+
+        Parameters
+        ----------
+        text : str
+            The text to classify
+
+        max_ngram_length : int
+            The maximum ngram length to use in classifying
+
+        Returns
+        -------
+        dict
+            {category:probability, category:probability...} or {} if no words
+            matching any categories in the model were found
+
+        """
+
+        model = self.weights
+        max_ngram_length = min(self.max_ngram_length, max_ngram_length)
+
+        raw_tokens = gptc.tokenizer.tokenize(
+            text, min(max_ngram_length, self.max_ngram_length)
         )
 
-    def serialize(self) -> bytes:
-        out = b"GPTC model v4\n"
-        out += (
+        tokens = gptc.tokenizer.hash_list(
+            raw_tokens,
+            self.hash_algorithm,
+        )
+
+        if transparent:
+            token_map = {tokens[i]: raw_tokens[i] for i in range(len(tokens))}
+            log: Log = []
+
+        numbered_probs: Dict[int, float] = {}
+
+        for word in tokens:
+            try:
+                unweighted_numbers = [
+                    i / 65535 for i in cast(List[float], model[word])
+                ]
+
+                weight, weighted_numbers = gptc.weighting.weight(
+                    unweighted_numbers
+                )
+
+                if transparent:
+                    log.append(
+                        (
+                            token_map[word],
+                            weight,
+                            unweighted_numbers,
+                        )
+                    )
+
+                for category, value in enumerate(weighted_numbers):
+                    try:
+                        numbered_probs[category] += value
+                    except KeyError:
+                        numbered_probs[category] = value
+            except KeyError:
+                pass
+
+        total = sum(numbered_probs.values())
+        probs: Dict[str, float] = {
+            self.names[category]: value / total
+            for category, value in numbered_probs.items()
+        }
+
+        if transparent:
+            explanation = convert_log(log, self.names)
+            return TransparentConfidences(probs, explanation)
+
+        return Confidences(probs)
+
+    def get(self, token: str) -> Dict[str, float]:
+        try:
+            weights = self.weights[
+                gptc.tokenizer.hash_single(
+                    gptc.tokenizer.normalize(token), self.hash_algorithm
+                )
+            ]
+        except KeyError:
+            return {}
+        return {
+            category: weights[index] / 65535
+            for index, category in enumerate(self.names)
+        }
+
+    def serialize(self, file: BinaryIO) -> None:
+        file.write(b"GPTC model v6\n")
+        file.write(
             json.dumps(
                 {
                     "names": self.names,
                     "max_ngram_length": self.max_ngram_length,
-                    "has_emoji": self.has_emoji,
+                    "hash_algorithm": self.hash_algorithm,
                 }
             ).encode("utf-8")
             + b"\n"
         )
         for word, weights in self.weights.items():
-            out += word.to_bytes(6, "big") + b"".join(
-                [weight.to_bytes(2, "big") for weight in weights]
+            file.write(
+                word.to_bytes(6, "big")
+                + b"".join([weight.to_bytes(2, "big") for weight in weights])
             )
-        return out
 
+    @staticmethod
+    def compile(
+        raw_model: Iterable[Mapping[str, str]],
+        max_ngram_length: int = 1,
+        min_count: int = 1,
+        hash_algorithm: str = "sha256",
+        ) -> 'Model':
+        """Compile a raw model.
 
-def deserialize(encoded_model: bytes) -> Model:
-    try:
-        prefix, config_json, encoded_weights = encoded_model.split(b"\n", 2)
-    except ValueError:
+        Parameters
+        ----------
+        raw_model : list of dict
+            A raw GPTC model.
+
+        max_ngram_length : int
+            Maximum ngram lenght to compile with.
+
+        Returns
+        -------
+        dict
+            A compiled GPTC model.
+
+        """
+        word_counts, category_lengths, names = _count_words(
+            raw_model, max_ngram_length, hash_algorithm
+        )
+        model = _get_weights(min_count, word_counts, category_lengths, names)
+        return Model(model, names, max_ngram_length, hash_algorithm)
+
+    @staticmethod
+    def deserialize(encoded_model: BinaryIO) -> "Model":
+        prefix = encoded_model.read(14)
+        if prefix != b"GPTC model v6\n":
             raise InvalidModelError()
 
-    if prefix != b"GPTC model v4":
+        config_json = b""
+        while True:
+            byte = encoded_model.read(1)
+            if byte == b"\n":
+                break
+
+            if byte == b"":
                 raise InvalidModelError()
 
+            config_json += byte
+
         try:
             config = json.loads(config_json.decode("utf-8"))
-    except (UnicodeDecodeError, json.JSONDecodeError):
-        raise InvalidModelError()
+        except (UnicodeDecodeError, json.JSONDecodeError) as exc:
+            raise InvalidModelError() from exc
 
         try:
             names = config["names"]
             max_ngram_length = config["max_ngram_length"]
-        has_emoji = config["has_emoji"]
-    except KeyError:
-        raise InvalidModelError()
+            hash_algorithm = config["hash_algorithm"]
+        except KeyError as exc:
+            raise InvalidModelError() from exc
 
         if not (
-        isinstance(names, list)
-        and isinstance(max_ngram_length, int)
-        and isinstance(has_emoji, bool)
-    ) or not all([isinstance(name, str) for name in names]):
+            isinstance(names, list) and isinstance(max_ngram_length, int)
+        ) or not all(isinstance(name, str) for name in names):
             raise InvalidModelError()
 
         weight_code_length = 6 + 2 * len(names)
 
-    if len(encoded_weights) % weight_code_length != 0:
+        weights: Dict[int, List[int]] = {}
+
+        while True:
+            code = encoded_model.read(weight_code_length)
+            if not code:
+                break
+            if len(code) != weight_code_length:
                 raise InvalidModelError()
 
-    weight_codes = [
-        encoded_weights[x : x + weight_code_length]
-        for x in range(0, len(encoded_weights), weight_code_length)
-    ]
-
-    weights = {
-        int.from_bytes(code[:6], "big"): [
+            weights[int.from_bytes(code[:6], "big")] = [
                 int.from_bytes(value, "big")
                 for value in [code[x : x + 2] for x in range(6, len(code), 2)]
             ]
-        for code in weight_codes
-    }
 
-    return Model(weights, names, max_ngram_length, has_emoji)
+        return Model(weights, names, max_ngram_length, hash_algorithm)

gptc/pack.py

@@ -7,7 +7,7 @@ from typing import List, Dict, Tuple
 
 def pack(
     directory: str, print_exceptions: bool = False
-) -> Tuple[List[Dict[str, str]], List[Tuple[Exception]]]:
+) -> Tuple[List[Dict[str, str]], List[Tuple[OSError]]]:
     paths = os.listdir(directory)
     texts: Dict[str, List[str]] = {}
     exceptions = []
@@ -17,16 +17,18 @@ def pack(
         try:
             for file in os.listdir(os.path.join(directory, path)):
                 try:
-                    with open(os.path.join(directory, path, file)) as f:
-                        texts[path].append(f.read())
-                except Exception as e:
-                    exceptions.append((e,))
+                    with open(
+                        os.path.join(directory, path, file), encoding="utf-8"
+                    ) as input_file:
+                        texts[path].append(input_file.read())
+                except OSError as error:
+                    exceptions.append((error,))
                     if print_exceptions:
-                        print(e, file=sys.stderr)
-        except Exception as e:
-            exceptions.append((e,))
+                        print(error, file=sys.stderr)
+        except OSError as error:
+            exceptions.append((error,))
             if print_exceptions:
-                print(e, file=sys.stderr)
+                print(error, file=sys.stderr)
 
     raw_model = []
 

gptc/tokenizer.py

@@ -1,25 +1,13 @@
 # SPDX-License-Identifier: GPL-3.0-or-later
 
-from typing import List, Union
+import unicodedata
+from typing import List, cast
 import hashlib
-import base64
-
-try:
-    import emoji
-
-    has_emoji = True
-except ImportError:
-    has_emoji = False
+import emoji
 
 
-def tokenize(
-    text: str, max_ngram_length: int = 1, use_emoji: bool = True
-) -> List[int]:
-    """Convert a string to a list of lemmas."""
-    converted_text: Union[str, List[str]] = text.lower()
-
-    if has_emoji and use_emoji:
-        text = text.lower()
+def tokenize(text: str, max_ngram_length: int = 1) -> List[str]:
+    text = unicodedata.normalize("NFKD", text).casefold()
     parts = []
     highest_end = 0
     for emoji_part in emoji.emoji_list(text):
@@ -32,9 +20,14 @@ def tokenize(
     tokens = [""]
 
     for char in converted_text:
-        if char.isalpha() or char == "'":
+        if (
+            char.isalpha()
+            or char.isnumeric()
+            or char == "'"
+            or (char in ",." and (" " + tokens[-1])[-1].isnumeric())
+        ):
             tokens[-1] += char
-        elif has_emoji and emoji.is_emoji(char):
+        elif emoji.is_emoji(char):
             tokens.append(char)
             tokens.append("")
         elif tokens[-1] != "":
@@ -43,16 +36,51 @@ def tokenize(
     tokens = [string for string in tokens if string]
 
     if max_ngram_length == 1:
-        ngrams = tokens
-    else:
+        return tokens
+
     ngrams = []
     for ngram_length in range(1, max_ngram_length + 1):
         for index in range(len(tokens) + 1 - ngram_length):
             ngrams.append(" ".join(tokens[index : index + ngram_length]))
+    return ngrams
 
-    return [
-        int.from_bytes(
-            hashlib.sha256(token.encode("utf-8")).digest()[:6], "big"
+
+def _hash_single(token: str, hash_function: type) -> int:
+    return int.from_bytes(
+        hash_function(token.encode("utf-8")).digest()[:6], "big"
     )
-        for token in ngrams
-    ]
+
+
+def _get_hash_function(hash_algorithm: str) -> type:
+    if hash_algorithm in {
+        "sha224",
+        "md5",
+        "sha512",
+        "sha3_256",
+        "blake2s",
+        "sha3_224",
+        "sha1",
+        "sha256",
+        "sha384",
+        "shake_256",
+        "blake2b",
+        "sha3_512",
+        "shake_128",
+        "sha3_384",
+    }:
+        return cast(type, getattr(hashlib, hash_algorithm))
+
+    raise ValueError("not a valid hash function: " + hash_algorithm)
+
+
+def hash_single(token: str, hash_algorithm: str) -> int:
+    return _hash_single(token, _get_hash_function(hash_algorithm))
+
+
+def hash_list(tokens: List[str], hash_algorithm: str) -> List[int]:
+    hash_function = _get_hash_function(hash_algorithm)
+    return [_hash_single(token, hash_function) for token in tokens]
+
+
+def normalize(text: str) -> str:
+    return " ".join(tokenize(text, 1))

gptc/weighting.py

@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: GPL-3.0-or-later
 
 import math
-from typing import Sequence, Union, Tuple, List
+from typing import Sequence, Tuple, List
 
 
 def _mean(numbers: Sequence[float]) -> float:
@@ -39,8 +39,8 @@ def _standard_deviation(numbers: Sequence[float]) -> float:
     return math.sqrt(_mean(squared_deviations))
 
 
-def weight(numbers: Sequence[float]) -> List[float]:
+def weight(numbers: Sequence[float]) -> Tuple[float, List[float]]:
     standard_deviation = _standard_deviation(numbers)
-    weight = standard_deviation * 2
-    weighted_numbers = [i * weight for i in numbers]
-    return weighted_numbers
+    weight_assigned = standard_deviation * 2
+    weighted_numbers = [i * weight_assigned for i in numbers]
+    return weight_assigned, weighted_numbers

profiler.py

@@ -0,0 +1,16 @@
+# SPDX-License-Identifier: GPL-3.0-or-later
+
+import cProfile
+import gptc
+import json
+import sys
+
+max_ngram_length = 10
+
+with open("models/raw.json") as f:
+    raw_model = json.load(f)
+
+with open("models/benchmark_text.txt") as f:
+    text = f.read()
+
+cProfile.run("gptc.Model.compile(raw_model, max_ngram_length)")

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gptc"
-version = "3.0.0"
+version = "4.0.1"
 description = "General-purpose text classifier"
 readme = "README.md"
 authors = [{ name = "Samuel Sloniker", email = "sam@kj7rrv.com"}]