Classifier objects will be removed in 5.0

A couple things I missed in 7f68dc6fc6

LGPL-3.0

@@ -1,165 +0,0 @@
-                   GNU LESSER GENERAL PUBLIC LICENSE
-                       Version 3, 29 June 2007
-
- Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
- Everyone is permitted to copy and distribute verbatim copies
- of this license document, but changing it is not allowed.
-
-
-  This version of the GNU Lesser General Public License incorporates
-the terms and conditions of version 3 of the GNU General Public
-License, supplemented by the additional permissions listed below.
-
-  0. Additional Definitions.
-
-  As used herein, "this License" refers to version 3 of the GNU Lesser
-General Public License, and the "GNU GPL" refers to version 3 of the GNU
-General Public License.
-
-  "The Library" refers to a covered work governed by this License,
-other than an Application or a Combined Work as defined below.
-
-  An "Application" is any work that makes use of an interface provided
-by the Library, but which is not otherwise based on the Library.
-Defining a subclass of a class defined by the Library is deemed a mode
-of using an interface provided by the Library.
-
-  A "Combined Work" is a work produced by combining or linking an
-Application with the Library.  The particular version of the Library
-with which the Combined Work was made is also called the "Linked
-Version".
-
-  The "Minimal Corresponding Source" for a Combined Work means the
-Corresponding Source for the Combined Work, excluding any source code
-for portions of the Combined Work that, considered in isolation, are
-based on the Application, and not on the Linked Version.
-
-  The "Corresponding Application Code" for a Combined Work means the
-object code and/or source code for the Application, including any data
-and utility programs needed for reproducing the Combined Work from the
-Application, but excluding the System Libraries of the Combined Work.
-
-  1. Exception to Section 3 of the GNU GPL.
-
-  You may convey a covered work under sections 3 and 4 of this License
-without being bound by section 3 of the GNU GPL.
-
-  2. Conveying Modified Versions.
-
-  If you modify a copy of the Library, and, in your modifications, a
-facility refers to a function or data to be supplied by an Application
-that uses the facility (other than as an argument passed when the
-facility is invoked), then you may convey a copy of the modified
-version:
-
-   a) under this License, provided that you make a good faith effort to
-   ensure that, in the event an Application does not supply the
-   function or data, the facility still operates, and performs
-   whatever part of its purpose remains meaningful, or
-
-   b) under the GNU GPL, with none of the additional permissions of
-   this License applicable to that copy.
-
-  3. Object Code Incorporating Material from Library Header Files.
-
-  The object code form of an Application may incorporate material from
-a header file that is part of the Library.  You may convey such object
-code under terms of your choice, provided that, if the incorporated
-material is not limited to numerical parameters, data structure
-layouts and accessors, or small macros, inline functions and templates
-(ten or fewer lines in length), you do both of the following:
-
-   a) Give prominent notice with each copy of the object code that the
-   Library is used in it and that the Library and its use are
-   covered by this License.
-
-   b) Accompany the object code with a copy of the GNU GPL and this license
-   document.
-
-  4. Combined Works.
-
-  You may convey a Combined Work under terms of your choice that,
-taken together, effectively do not restrict modification of the
-portions of the Library contained in the Combined Work and reverse
-engineering for debugging such modifications, if you also do each of
-the following:
-
-   a) Give prominent notice with each copy of the Combined Work that
-   the Library is used in it and that the Library and its use are
-   covered by this License.
-
-   b) Accompany the Combined Work with a copy of the GNU GPL and this license
-   document.
-
-   c) For a Combined Work that displays copyright notices during
-   execution, include the copyright notice for the Library among
-   these notices, as well as a reference directing the user to the
-   copies of the GNU GPL and this license document.
-
-   d) Do one of the following:
-
-       0) Convey the Minimal Corresponding Source under the terms of this
-       License, and the Corresponding Application Code in a form
-       suitable for, and under terms that permit, the user to
-       recombine or relink the Application with a modified version of
-       the Linked Version to produce a modified Combined Work, in the
-       manner specified by section 6 of the GNU GPL for conveying
-       Corresponding Source.
-
-       1) Use a suitable shared library mechanism for linking with the
-       Library.  A suitable mechanism is one that (a) uses at run time
-       a copy of the Library already present on the user's computer
-       system, and (b) will operate properly with a modified version
-       of the Library that is interface-compatible with the Linked
-       Version.
-
-   e) Provide Installation Information, but only if you would otherwise
-   be required to provide such information under section 6 of the
-   GNU GPL, and only to the extent that such information is
-   necessary to install and execute a modified version of the
-   Combined Work produced by recombining or relinking the
-   Application with a modified version of the Linked Version. (If
-   you use option 4d0, the Installation Information must accompany
-   the Minimal Corresponding Source and Corresponding Application
-   Code. If you use option 4d1, you must provide the Installation
-   Information in the manner specified by section 6 of the GNU GPL
-   for conveying Corresponding Source.)
-
-  5. Combined Libraries.
-
-  You may place library facilities that are a work based on the
-Library side by side in a single library together with other library
-facilities that are not Applications and are not covered by this
-License, and convey such a combined library under terms of your
-choice, if you do both of the following:
-
-   a) Accompany the combined library with a copy of the same work based
-   on the Library, uncombined with any other library facilities,
-   conveyed under the terms of this License.
-
-   b) Give prominent notice with the combined library that part of it
-   is a work based on the Library, and explaining where to find the
-   accompanying uncombined form of the same work.
-
-  6. Revised Versions of the GNU Lesser General Public License.
-
-  The Free Software Foundation may publish revised and/or new versions
-of the GNU Lesser General Public License from time to time. Such new
-versions will be similar in spirit to the present version, but may
-differ in detail to address new problems or concerns.
-
-  Each version is given a distinguishing version number. If the
-Library as you received it specifies that a certain numbered version
-of the GNU Lesser General Public License "or any later version"
-applies to it, you have the option of following the terms and
-conditions either of that published version or of any later version
-published by the Free Software Foundation. If the Library as you
-received it does not specify a version number of the GNU Lesser
-General Public License, you may choose any version of the GNU Lesser
-General Public License ever published by the Free Software Foundation.
-
-  If the Library as you received it specifies that a proxy can decide
-whether future versions of the GNU Lesser General Public License shall
-apply, that proxy's public statement of acceptance of any version is
-permanent authorization for you to choose that version for the
-Library.

LICENSE

@@ -1,14 +1,13 @@
 Copyright (c) 2020-2022 Samuel L Sloniker
 
 This program is free software: you can redistribute it and/or modify it under
-the terms of the GNU Lesser General Public License as published by the Free
-Software Foundation, either version 3 of the License, or (at your option) any
-later version.
+the terms of the GNU General Public License as published by the Free Software
+Foundation, either version 3 of the License, or (at your option) any later
+version.
 
 This program is distributed in the hope that it will be useful, but WITHOUT ANY
 WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A
 PARTICULAR PURPOSE. See the GNU General Public License for more details.
 
-You should have received copies of the GNU General Public License and the GNU
-Lesser General Public License along with this program. If not, see
-<https://www.gnu.org/licenses/>. 
+You should have received a copy of the GNU General Public License along with
+this program. If not, see <https://www.gnu.org/licenses/>. 

README.md

@@ -4,78 +4,142 @@ General-purpose text classifier in Python
 
 GPTC provides both a CLI tool and a Python library.
 
+## Installation
+
+    pip install gptc
+
 ## CLI Tool
 
 ### Classifying text
 
-    python -m gptc classify [-n <max_ngram_length>] <compiled model file>
+    gptc classify [-n <max_ngram_length>] <compiled model file>
 
 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
 
-    python -m 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
 
-    python -m gptc compile [-n <max_ngram_length>] <raw model file>
+    gptc compile [-n <max_ngram_length>] [-c <min_count>] <raw model file> <compiled model file>
 
-This will print the compiled model in JSON 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.
+
+### Packing models
+
+    gptc pack <dir>
+
+This will print the raw model in JSON to stdout. See `models/unpacked/` for an
+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 dict, not
-JSON).
+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, ...}`
 
-#### `Classifier.classify(text)`
-
-Classify `text`. Returns the category into which the text is placed (as a
-string), or `None` when it cannot classify the text.
-
-### `gptc.compile(raw_model, max_ngram_length=1)`
-Compile a raw model (as a list, not JSON) and return the compiled model (as a
-dict).
+Note that this may not include values for all categories. If there are no
+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.
 
 For information about `max_ngram_length`, see section "Ngrams."
 
+### `Model.get(token)`
+
+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.
+
+### `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).
+
+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.
+
+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:
+
+* `md5`
+* `sha1`
+* `sha224`
+* `sha256`
+* `sha384`
+* `sha512`
+* `sha3_224`
+* `sha3_384`
+* `sha3_256`
+* `sha3_512`
+* `shake_128`
+* `shake_256`
+* `blake2b`
+* `blake2s`
+
+### `gptc.pack(directory, print_exceptions=False)`
+
+Pack the model in `directory` and return a tuple of the format:
+
+    (raw_model, [(exception,),(exception,)...])
+
+Note that the exceptions are contained in single-item tuples. This is to allow
+more information to be provided without breaking the API in future versions of
+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
-are disabled by default (maximum length set to 1) for performance and
-compatibility reasons. Enabling them significantly increases the time required
-both for compilation and classification. The effect seems more significant for
-compilation than for classification. Compiled models are also much larger when
-ngrams are enabled. Larger maximum ngram lengths will result in slower
-performance and larger files. It is a good idea to experiment with different
-values and use the highest one at which GPTC is fast enough and models are
-small enough for your needs.
+are disabled by default (maximum length set to 1) for performance reasons.
+Enabling them significantly increases the time required both for compilation
+and classification. The effect seems more significant for compilation than for
+classification. Compiled models are also much larger when ngrams are enabled.
+Larger maximum ngram lengths will result in slower performance and larger
+files. It is a good idea to experiment with different values and use the
+highest one at which GPTC is fast enough and models are small enough for your
+needs.
 
 Once a model is compiled at a certain maximum ngram length, it cannot be used
 for classification with a higher value. If you instantiate a `Classifier` with
 a model compiled with a lower `max_ngram_length`, the value will be silently
 reduced to the one used when compiling the model.
 
-Models compiled with older versions of GPTC which did not support ngrams are
-handled the same way as models compiled with `max_ngram_length=1`.
-
 ## Model format
 
-This section explains the raw model format, which is how you should create and
-edit models.
+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:
 
@@ -86,10 +150,14 @@ Raw models are formatted as a list of dicts. See below for the format:
         }
     ]
 
-GPTC handles models as Python `list`s of `dict`s of `str`s (for raw models) or
-`dict`s of `str`s and `float`s (for compiled models), 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
+
+GPTC treats individual emoji as words.
 
 ## Example model
 

benchmark.py

@@ -1,3 +1,5 @@
+# SPDX-License-Identifier: GPL-3.0-or-later
+
 import timeit
 import gptc
 import json
@@ -23,7 +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(),
         )
@@ -33,7 +35,9 @@ print(
 )
 
 
-classifier = gptc.Classifier(gptc.compile(raw_model, max_ngram_length), max_ngram_length)
+classifier = gptc.Classifier(
+    gptc.compile(raw_model, max_ngram_length), max_ngram_length
+)
 print(
     "Average classification time over",
     classify_iterations,

gptc/__init__.py

@@ -1,7 +1,12 @@
-# SPDX-License-Identifier: LGPL-3.0-or-later
+# SPDX-License-Identifier: GPL-3.0-or-later
 
 """General-Purpose Text Classifier"""
 
-from gptc.compiler import compile
-from gptc.classifier import Classifier
-from gptc.exceptions import *
+from gptc.pack import pack
+from gptc.model import Model
+from gptc.tokenizer import normalize
+from gptc.exceptions import (
+    GPTCError,
+    ModelError,
+    InvalidModelError,
+)

gptc/__main__.py

@@ -1,57 +1,87 @@
 #!/usr/bin/env python3
-# SPDX-License-Identifier: LGPL-3.0-or-later
+# SPDX-License-Identifier: GPL-3.0-or-later
 
 import argparse
 import json
 import sys
 import gptc
 
-def main():
+
+def main() -> None:
     parser = argparse.ArgumentParser(
         description="General Purpose Text Classifier", prog="gptc"
     )
     subparsers = parser.add_subparsers(dest="subparser_name", required=True)
 
-    compile_parser = subparsers.add_parser("compile", help="compile a raw model")
+    compile_parser = subparsers.add_parser(
+        "compile", help="compile a raw model"
+    )
     compile_parser.add_argument("model", help="raw model to compile")
-    compile_parser.add_argument("--max-ngram-length", "-n", help="maximum ngram length", type=int, default=1)
+    compile_parser.add_argument(
+        "out", help="name of file to write compiled model to"
+    )
+    compile_parser.add_argument(
+        "--max-ngram-length",
+        "-n",
+        help="maximum ngram length",
+        type=int,
+        default=1,
+    )
+    compile_parser.add_argument(
+        "--min-count",
+        "-c",
+        help="minimum use count for word/ngram to be included in model",
+        type=int,
+        default=1,
+    )
 
     classify_parser = subparsers.add_parser("classify", help="classify text")
     classify_parser.add_argument("model", help="compiled model to use")
-    classify_parser.add_argument("--max-ngram-length", "-n", help="maximum ngram length", 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",
+    classify_parser.add_argument(
+        "--max-ngram-length",
+        "-n",
+        help="maximum ngram length",
+        type=int,
+        default=1,
     )
-    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"
+    )
+    pack_parser.add_argument("model", help="directory containing model")
 
     args = parser.parse_args()
 
-    with open(args.model, "r") as f:
-        model = json.load(f)
-
     if args.subparser_name == "compile":
-        print(json.dumps(gptc.compile(model, args.max_ngram_length)))
-    else:
-        classifier = gptc.Classifier(model, args.max_ngram_length)
+        with open(args.model, "r", encoding="utf-8") as input_file:
+            model = json.load(input_file)
+
+        with open(args.out, "wb+") as output_file:
+            gptc.Model.compile(
+                model, args.max_ngram_length, args.min_count
+            ).serialize(output_file)
+    elif args.subparser_name == "classify":
+        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]))
 
 
 if __name__ == "__main__":

gptc/classifier.py

@@ -1,96 +0,0 @@
-# SPDX-License-Identifier: LGPL-3.0-or-later
-
-import gptc.tokenizer, gptc.compiler, gptc.exceptions, gptc.weighting
-import warnings
-
-
-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, max_ngram_length=1):
-        if model.get("__version__", 0) != 3:
-            raise gptc.exceptions.UnsupportedModelError(
-                f"unsupported model version"
-            )
-        self.model = model
-        self.max_ngram_length = min(
-            max_ngram_length, model.get("__ngrams__", 1)
-        )
-
-    def confidence(self, text):
-        """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
-
-        text = gptc.tokenizer.tokenize(text, self.max_ngram_length)
-        probs = {}
-        for word in text:
-            try:
-                weight, weighted_numbers = gptc.weighting.weight(
-                    [i / 65535 for i in model[word]]
-                )
-                for category, value in enumerate(weighted_numbers):
-                    try:
-                        probs[category] += value
-                    except KeyError:
-                        probs[category] = value
-            except KeyError:
-                pass
-        probs = {
-            model["__names__"][category]: value
-            for category, value in probs.items()
-        }
-        total = sum(probs.values())
-        probs = {category: value / total for category, value in probs.items()}
-        return probs
-
-    def classify(self, text):
-        """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 = self.confidence(text)
-        try:
-            return sorted(probs.items(), key=lambda x: x[1])[-1][0]
-        except IndexError:
-            return None

gptc/compiler.py

@@ -1,73 +0,0 @@
-# SPDX-License-Identifier: LGPL-3.0-or-later
-
-import gptc.tokenizer
-
-
-def compile(raw_model, max_ngram_length=1):
-    """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 = {}
-
-    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
-
-    categories_by_count = {}
-
-    names = []
-
-    for category, text in categories.items():
-        if not category in names:
-            names.append(category)
-
-        categories_by_count[category] = {}
-        for word in text:
-            try:
-                categories_by_count[category][word] += 1 / len(
-                    categories[category]
-                )
-            except KeyError:
-                categories_by_count[category][word] = 1 / len(
-                    categories[category]
-                )
-    word_weights = {}
-    for category, words in categories_by_count.items():
-        for word, value in words.items():
-            try:
-                word_weights[word][category] = value
-            except KeyError:
-                word_weights[word] = {category: value}
-
-    model = {}
-    for word, weights in word_weights.items():
-        total = sum(weights.values())
-        model[word] = []
-        for category in names:
-            model[word].append(
-                round((weights.get(category, 0) / total) * 65535)
-            )
-
-    model["__names__"] = names
-    model["__ngrams__"] = max_ngram_length
-    model["__version__"] = 3
-
-    return model

gptc/exceptions.py

@@ -1,4 +1,4 @@
-# SPDX-License-Identifier: LGPL-3.0-or-later
+# SPDX-License-Identifier: GPL-3.0-or-later
 
 
 class GPTCError(BaseException):
@@ -9,5 +9,5 @@ class ModelError(GPTCError):
     pass
 
 
-class UnsupportedModelError(ModelError):
+class InvalidModelError(ModelError):
     pass

gptc/model.py

@@ -0,0 +1,322 @@
+# 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
+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:
+    def __init__(
+        self,
+        weights: Dict[int, List[int]],
+        names: List[str],
+        max_ngram_length: int,
+        hash_algorithm: str,
+    ):
+        self.weights = weights
+        self.names = names
+        self.max_ngram_length = max_ngram_length
+        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)
+        )
+
+        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,
+                    "hash_algorithm": self.hash_algorithm,
+                }
+            ).encode("utf-8")
+            + b"\n"
+        )
+        for word, weights in self.weights.items():
+            file.write(
+                word.to_bytes(6, "big")
+                + b"".join([weight.to_bytes(2, "big") for weight in weights])
+            )
+
+    @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.
+
+        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()
+
+        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) as exc:
+            raise InvalidModelError() from exc
+
+        try:
+            names = config["names"]
+            max_ngram_length = config["max_ngram_length"]
+            hash_algorithm = config["hash_algorithm"]
+        except KeyError as exc:
+            raise InvalidModelError() from exc
+
+        if not (
+            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)
+
+        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()
+
+            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)]
+            ]
+
+        return Model(weights, names, max_ngram_length, hash_algorithm)

gptc/pack.py

@@ -0,0 +1,38 @@
+# SPDX-License-Identifier: GPL-3.0-or-later
+
+import sys
+import os
+from typing import List, Dict, Tuple
+
+
+def pack(
+    directory: str, print_exceptions: bool = False
+) -> Tuple[List[Dict[str, str]], List[Tuple[OSError]]]:
+    paths = os.listdir(directory)
+    texts: Dict[str, List[str]] = {}
+    exceptions = []
+
+    for path in paths:
+        texts[path] = []
+        try:
+            for file in os.listdir(os.path.join(directory, path)):
+                try:
+                    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(error, file=sys.stderr)
+        except OSError as error:
+            exceptions.append((error,))
+            if print_exceptions:
+                print(error, file=sys.stderr)
+
+    raw_model = []
+
+    for category, cat_texts in texts.items():
+        raw_model += [{"category": category, "text": i} for i in cat_texts]
+
+    return raw_model, exceptions

gptc/tokenizer.py

@@ -1,13 +1,35 @@
-# SPDX-License-Identifier: LGPL-3.0-or-later
+# SPDX-License-Identifier: GPL-3.0-or-later
+
+import unicodedata
+from typing import List, cast
+import hashlib
+import emoji
 
 
-def tokenize(text, max_ngram_length=1):
-    """Convert a string to a list of lemmas."""
+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):
+        parts += list(text[highest_end : emoji_part["match_start"]])
+        parts.append(emoji_part["emoji"])
+        highest_end = emoji_part["match_end"]
+    parts += list(text[highest_end:])
+    converted_text = [part for part in parts if part]
+
     tokens = [""]
 
-    for char in text.lower():
-        if char.isalpha() or char == "'":
+    for char in converted_text:
+        if (
+            char.isalpha()
+            or char.isnumeric()
+            or char == "'"
+            or (char in ",." and (" " + tokens[-1])[-1].isnumeric())
+        ):
             tokens[-1] += char
+        elif emoji.is_emoji(char):
+            tokens.append(char)
+            tokens.append("")
         elif tokens[-1] != "":
             tokens.append("")
 
@@ -15,9 +37,50 @@ def tokenize(text, max_ngram_length=1):
 
     if max_ngram_length == 1:
         return tokens
-    else:
-        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
+
+    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
+
+
+def _hash_single(token: str, hash_function: type) -> int:
+    return int.from_bytes(
+        hash_function(token.encode("utf-8")).digest()[:6], "big"
+    )
+
+
+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

@@ -1,9 +1,10 @@
-# SPDX-License-Identifier: LGPL-3.0-or-later
+# SPDX-License-Identifier: GPL-3.0-or-later
 
 import math
+from typing import Sequence, Tuple, List
 
 
-def _mean(numbers):
+def _mean(numbers: Sequence[float]) -> float:
     """Calculate the mean of a group of numbers
 
     Parameters
@@ -19,7 +20,7 @@ def _mean(numbers):
     return sum(numbers) / len(numbers)
 
 
-def _standard_deviation(numbers):
+def _standard_deviation(numbers: Sequence[float]) -> float:
     """Calculate the standard deviation of a group of numbers
 
     Parameters
@@ -38,8 +39,8 @@ def _standard_deviation(numbers):
     return math.sqrt(_mean(squared_deviations))
 
 
-def weight(numbers):
+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 weight, 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 = "2.0.1"
+version = "4.0.1"
 description = "General-purpose text classifier"
 readme = "README.md"
 authors = [{ name = "Samuel Sloniker", email = "sam@kj7rrv.com"}]
@@ -12,10 +12,10 @@ classifiers = [
     "Programming Language :: Python",
     "Programming Language :: Python :: 3",
     "Development Status :: 5 - Production/Stable",
-    "License :: OSI Approved :: GNU Lesser General Public License v3 or later (LGPLv3+)",
+    "License :: OSI Approved :: GNU General Public License v3 or later (GPLv3+)",
     "Operating System :: OS Independent",
 ]
-dependencies = []
+dependencies = ["emoji"]
 requires-python = ">=3.7"
 
 [project.urls]

utils/pack.py

@@ -1,41 +0,0 @@
-# SPDX-License-Identifier: LGPL-3.0-or-later
-
-import sys
-import os
-import json
-
-
-def pack(directory, print_exceptions=True):
-    paths = os.listdir(directory)
-    texts = {}
-    exceptions = []
-
-    for path in paths:
-        texts[path] = []
-        try:
-            for file in os.listdir(os.path.join(sys.argv[1], path)):
-                try:
-                    with open(os.path.join(sys.argv[1], path, file)) as f:
-                        texts[path].append(f.read())
-                except Exception as e:
-                    exceptions.append((e,))
-                    if print_exceptions:
-                        print(e, file=sys.stderr)
-        except Exception as e:
-            exceptions.append((e,))
-            if print_exceptions:
-                print(e, file=sys.stderr)
-
-    raw_model = []
-
-    for category, cat_texts in texts.items():
-        raw_model += [{"category": category, "text": i} for i in cat_texts]
-
-    return raw_model, exceptions
-
-
-if len(sys.argv) != 2:
-    print("usage: pack.py <path>", file=sys.stderr)
-    exit(1)
-
-print(json.dumps(pack(sys.argv[1])[0]))