hdx-python-utilities 4.1.1__tar.gz → 4.2.0__tar.gz

{hdx_python_utilities-4.1.1 → hdx_python_utilities-4.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.5
 Name: hdx-python-utilities
-Version: 4.1.1
+Version: 4.2.0
 Summary: HDX Python Utilities for streaming tabular data, date and time handling and other helpful functions
 Project-URL: Homepage, https://github.com/OCHA-DAP/hdx-python-utilities
 Author-email: Michael Rans <rans@email.com>
@@ -32,6 +32,7 @@ Requires-Dist: loguru
 Requires-Dist: openpyxl>=3.1.2
 Requires-Dist: pyphonetics
 Requires-Dist: python-dateutil<2.9.1,>=2.9.0
+Requires-Dist: rapidfuzz
 Requires-Dist: ratelimit
 Requires-Dist: requests-file
 Requires-Dist: ruamel-yaml

{hdx_python_utilities-4.1.1 → hdx_python_utilities-4.2.0}/documentation/index.md

@@ -33,6 +33,11 @@ The code for the library is [here](https://github.com/OCHA-DAP/hdx-python-utilit
 The library has detailed API documentation which can be found in the menu at the top.
 
 ## Breaking Changes
+From 4.2.0, get_code_from_name fuzzy matches using RapidFuzzMatcher rather than
+Phonetics by default. Pass matcher=Phonetics() for the previous behaviour. A name
+made of parts (eg. "Protection - GBV") is resolved from its parts, and a list of
+names (eg. "Wash & Protection") is not fuzzy matched.
+
 From 4.1.0, save_hxlated_output and Download.hxl_row removed. Parameter use_hxl removed
 from all calls in Download and Retrieve.
 
@@ -205,7 +210,15 @@ Other useful functions:
     # if no folder supplied
     # path = Download.get_path_for_url(url, folder)
 
-For more detail and additional functions, check the API documentation.
+### Resumable downloads
+
+Pass `resume=True` and a fixed `path` to continue an interrupted download using
+HTTP Range requests. If the server returns 200 instead of 206, the file is
+downloaded from scratch. Add `retries` to automatically retry mid-stream
+failures; each retry continues from the current end of the partial file.
+`retries` has no effect without `resume=True`.
+
+    path = downloader.download_file(url, path=mypath, resume=True, retries=3)
 
 ## Retrieving files
 
@@ -261,6 +274,8 @@ Examples:
         retriever = Retrieve(downloader, fallback_dir, saved_dir, temp_dir, save=False, use_saved=True)
         data = retriever.download_json(url, filename, logstr="test json", fallback=False, log_level=logging.DEBUG)
 
+`Retrieve.download_file` also accepts `resume=True` and `retries=N` — see [Resumable downloads](#resumable-downloads).
+
 ## Date utilities
 
 There are utilities to parse dates. By default, *no timezone information will be parsed
@@ -867,7 +882,19 @@ Efficient Hashing of files that produces a stable hash even for zip and xlsx fil
 
 Examples:
 
+    # Fuzzy match a name to one of a list of names using rapidfuzz scoring
     possible_names = ["Al Maharah", "Ad Dali", "Dhamar"]
+    matcher = RapidFuzzMatcher()
+    assert matcher.match(possible_names, "al dali") == 1
+
+    # place_name_scorer does not match place names that share a word but differ
+    # in another that distinguishes places
+    kalimantan = ["kalimantan utara", "kalimantan tengah"]
+    assert matcher.match(kalimantan, "central kalimantan") == 0
+    matcher = RapidFuzzMatcher(RapidFuzzMatcher.place_name_scorer)
+    assert matcher.match(kalimantan, "central kalimantan") is None
+
+    # Phonetics is an alternative matcher using Refined Soundex
     phonetics = Phonetics()
     assert phonetics.match(possible_names, "al dali") == 1
 
@@ -877,6 +904,22 @@ Examples:
     }
     assert get_code_from_name("NATIONAL_NGO", lookup, [], fuzzy_match=False) == "441"
 
+    # get_code_from_name fuzzy matches with RapidFuzzMatcher unless given
+    # another matcher. It resolves names made of parts and does not fuzzy
+    # match lists of names.
+    sector_lookup = {"food security": "FSC", "protection": "PRO", "gbv": "PRO-GBV", "wash": "WSH"}
+    assert get_code_from_name("Food Securty", sector_lookup, []) == "FSC"
+    assert get_code_from_name("Food Securty", sector_lookup, [], matcher=Phonetics()) == "FSC"
+    assert get_code_from_name("Protection - GBV", sector_lookup, []) == "PRO-GBV"
+    assert get_code_from_name("Wash & Protection", sector_lookup, []) is None
+
+    # Resolve a name made of parts, returning the code and whether it is a
+    # list of names. The third parameter is a part that qualifies the other,
+    # eg. a parent admin name.
+    admin_lookup = {"acosta": "VE1101", "falcon": "VE1109"}
+    assert resolve_name_parts("Falcón, Acosta", admin_lookup) == (None, True)
+    assert resolve_name_parts("Falcón, Acosta", admin_lookup, "falcon") == ("VE1101", False)
+
     a = "The quick brown fox jumped over the lazy dog. It was so fast!"
 
     # Replace multiple strings in a string simultaneously

{hdx_python_utilities-4.1.1 → hdx_python_utilities-4.2.0}/pyproject.toml

@@ -52,6 +52,7 @@ dependencies = [
     "typing_extensions",
     "loguru",
     "pyphonetics",
+    "rapidfuzz",
     "python-dateutil>=2.9.0, <2.9.1",
     "ratelimit",
     "requests-file",
@@ -70,6 +71,7 @@ dev = [
     "pytest",
     "pytest-cov",
     "pytest-loguru",
+    "pytest-mock",
     "pre-commit",
     "ruff==0.14.13",
 ]

{hdx_python_utilities-4.1.1 → hdx_python_utilities-4.2.0}/src/hdx/utilities/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '4.1.1'
-__version_tuple__ = version_tuple = (4, 1, 1)
+__version__ = version = '4.2.0'
+__version_tuple__ = version_tuple = (4, 2, 0)
 
 __commit_id__ = commit_id = None

{hdx_python_utilities-4.1.1 → hdx_python_utilities-4.2.0}/src/hdx/utilities/downloader.py

@@ -250,13 +250,16 @@ class Download(BaseDownload):
                 f"Download of {url} failed in retrieval of stream!" % url
             )
 
-    def stream_path(self, path: Path | str, errormsg: str) -> Path:
+    def stream_path(
+        self, path: Path | str, errormsg: str, append: bool = False
+    ) -> Path:
         """Stream file from url and store in provided path. Must call setup
         method first.
 
         Args:
             path: Path for downloaded file
             errormsg: Error message to display if there is a problem
+            append: Whether to append to an existing file. Defaults to False.
 
         Returns:
             Path of downloaded file
@@ -264,7 +267,7 @@ class Download(BaseDownload):
         f = None
         try:
             path = Path(path)
-            f = path.open("wb")
+            f = path.open("ab" if append else "wb")
             for chunk in self.response.iter_content(chunk_size=10240):
                 if chunk:  # filter out keep-alive new chunks
                     f.write(chunk)
@@ -325,6 +328,8 @@ class Download(BaseDownload):
             path (str): Full path to use for downloaded file instead of folder and filename.
             overwrite (bool): Whether to overwrite existing file. Defaults to False.
             keep (bool): Whether to keep already downloaded file. Defaults to False.
+            resume (bool): Whether to resume a partial download using Range requests where the server supports it. Defaults to False.
+            retries (int): Number of times to retry a mid-stream failure. Only effective when resume=True so that each retry can continue from the partial file. Defaults to 0.
             post (bool): Whether to use POST instead of GET. Defaults to False.
             parameters (dict): Parameters to pass. Defaults to None.
             timeout (float): Timeout for connecting to URL. Defaults to None (no timeout).
@@ -340,25 +345,84 @@ class Download(BaseDownload):
         path = kwargs.get("path")
         overwrite = kwargs.get("overwrite", False)
         keep = kwargs.get("keep", False)
+        resume = kwargs.get("resume", False)
+        retries = kwargs.get("retries", 0)
         try:
-            path = get_path_for_url(url, folder, filename, path, overwrite, keep)
+            # When resuming, skip uniqueness renaming so we get back the exact target path
+            path = get_path_for_url(
+                url, folder, filename, path, overwrite, keep or resume
+            )
         except ValueError as ex:
             raise DownloadError(ex) from ex
-        if keep and exists(path):
+        if keep and not resume and exists(path):
             return path
-        self.setup(
-            url,
-            stream=True,
-            post=kwargs.get("post", False),
-            parameters=kwargs.get("parameters"),
-            timeout=kwargs.get("timeout"),
-            headers=kwargs.get("headers"),
-            encoding=kwargs.get("encoding"),
-            json_string=kwargs.get("json_string", False),
-        )
-        return self.stream_path(
-            path, f"Download of {url} failed in retrieval of stream!"
-        )
+        errormsg = f"Download of {url} failed in retrieval of stream!"
+        setup_headers = dict(kwargs.get("headers") or {})
+        append = False
+        if resume and path.exists() and path.stat().st_size > 0:
+            setup_headers["Range"] = f"bytes={path.stat().st_size}-"
+            # Range bytes must come from the raw stream: partial gzip streams
+            # lack the header and cannot be decoded, so request uncompressed.
+            setup_headers.setdefault("Accept-Encoding", "identity")
+            try:
+                self.setup(
+                    url,
+                    stream=True,
+                    post=kwargs.get("post", False),
+                    parameters=kwargs.get("parameters"),
+                    timeout=kwargs.get("timeout"),
+                    headers=setup_headers,
+                    encoding=kwargs.get("encoding"),
+                    json_string=kwargs.get("json_string", False),
+                )
+            except DownloadError:
+                if self.response is not None and self.response.status_code == 416:
+                    return path
+                raise
+            append = self.response.status_code == 206
+        else:
+            self.setup(
+                url,
+                stream=True,
+                post=kwargs.get("post", False),
+                parameters=kwargs.get("parameters"),
+                timeout=kwargs.get("timeout"),
+                headers=setup_headers or None,
+                encoding=kwargs.get("encoding"),
+                json_string=kwargs.get("json_string", False),
+            )
+        attempts_remaining = retries
+        while True:
+            try:
+                return self.stream_path(path, errormsg, append=append)
+            except DownloadError:
+                if not resume or attempts_remaining == 0:
+                    raise
+                attempts_remaining -= 1
+                logger.warning(
+                    "Download of %s interrupted, retrying (attempt %d of %d)...",
+                    url,
+                    retries - attempts_remaining,
+                    retries,
+                )
+                setup_headers["Range"] = f"bytes={path.stat().st_size}-"
+                setup_headers.setdefault("Accept-Encoding", "identity")
+                try:
+                    self.setup(
+                        url,
+                        stream=True,
+                        post=kwargs.get("post", False),
+                        parameters=kwargs.get("parameters"),
+                        timeout=kwargs.get("timeout"),
+                        headers=setup_headers,
+                        encoding=kwargs.get("encoding"),
+                        json_string=kwargs.get("json_string", False),
+                    )
+                except DownloadError:
+                    if self.response is not None and self.response.status_code == 416:
+                        return path
+                    raise
+                append = self.response.status_code == 206
 
     def download(self, url: Path | str, **kwargs: Any) -> requests.Response:
         """Download url.

{hdx_python_utilities-4.1.1 → hdx_python_utilities-4.2.0}/src/hdx/utilities/matching.py

@@ -3,10 +3,14 @@ import re
 from collections.abc import Callable, Sequence
 
 from pyphonetics import RefinedSoundex
+from rapidfuzz import fuzz as rapidfuzz_fuzz
 
 from hdx.utilities.text import normalise
 
 TEMPLATE_VARIABLES = re.compile("{{.*?}}")
+NON_WORD_CHARACTERS = re.compile(r"[\W_]+")
+LIST_SEPARATORS = re.compile(r"\s*(?:[,;/:&\n]|\s(?:and|et)\s)\s*", re.IGNORECASE)
+HYPHEN = re.compile(r"\s*-\s*")
 
 
 class Phonetics(RefinedSoundex):
@@ -58,15 +62,212 @@ class Phonetics(RefinedSoundex):
         return matching_index
 
 
+class RapidFuzzMatcher:
+    """
+    Default matcher for names, using rapidfuzz edit-distance scoring instead
+    of phonetic (Soundex-family) encoding as Phonetics does. Phonetic algorithms
+    encode English pronunciation rules, so they are a poor fit for matching
+    multilingual or transliterated names (eg. "Sana'a" vs "Sanaa") that are
+    spelling variants rather than being phonetically similar in English
+    terms. Edit distance on normalised strings tends to place such variants
+    closer together.
+
+    Args:
+        scorer: Function taking two strings and returning a similarity score
+            from 0-100. Defaults to default_scorer. Use place_name_scorer for
+            place names. Ties are broken by rapidfuzz.fuzz.ratio over the
+            whole strings.
+
+    Caveat: "<name> city" scores ~100 against "<name>", as for any name whose
+    words are a subset of the other's. So withholding a text replacement that
+    strips " city" does not stop "Kenge city" matching "Kenge".
+    """
+
+    def __init__(self, scorer: Callable[[str, str], float] | None = None) -> None:
+        self.scorer = scorer or self.default_scorer
+
+    @staticmethod
+    def default_scorer(name: str, possible_name: str) -> float:
+        """
+        Score two strings with token_set_ratio, or with ratio on the strings
+        with spaces and punctuation removed if that is at least 90 and higher.
+
+        Args:
+            name: Name to match
+            possible_name: Possible name
+
+        Returns:
+            Similarity score from 0-100
+        """
+        score = rapidfuzz_fuzz.token_set_ratio(name, possible_name)
+        joined_score = rapidfuzz_fuzz.ratio(
+            NON_WORD_CHARACTERS.sub("", name),
+            NON_WORD_CHARACTERS.sub("", possible_name),
+        )
+        if joined_score >= 90:
+            return max(score, joined_score)
+        return score
+
+    @staticmethod
+    def place_name_scorer(name: str, possible_name: str) -> float:
+        """
+        Score two strings with default_scorer, but score 0 if they share a
+        word, each has other words of 3+ characters, and one of the words on
+        the side with fewer has ratio < 60 to all those on the other side (eg.
+        "central kalimantan" vs "kalimantan utara", but not "central west" vs
+        "centre west"). Suits place names, where the differing word usually
+        distinguishes places, but not names like sectors, where extra words
+        are often descriptive (eg. "food security and livelihood").
+
+        Args:
+            name: Name to match
+            possible_name: Possible name
+
+        Returns:
+            Similarity score from 0-100
+        """
+        words = set(name.split())
+        possible_words = set(possible_name.split())
+
+        def long_words(words: set[str]) -> set[str]:
+            return {word for word in words if len(word) > 2}
+
+        extra_words = long_words(words - possible_words)
+        possible_extra_words = long_words(possible_words - words)
+        if words & possible_words and extra_words and possible_extra_words:
+            fewer, more = sorted((extra_words, possible_extra_words), key=len)
+            for word in fewer:
+                if max(rapidfuzz_fuzz.ratio(word, other) for other in more) < 60:
+                    return 0.0
+        return RapidFuzzMatcher.default_scorer(name, possible_name)
+
+    def match(
+        self,
+        possible_names: Sequence,
+        name: str,
+        alternative_name: str | None = None,
+        transform_possible_names: Sequence[Callable] = [],
+        threshold: float = 65.0,
+    ) -> int | None:
+        """
+        Match name to one of the given possible names. Returns None if no
+        match or the index of the matching name.
+
+        Args:
+            possible_names: Possible names
+            name: Name to match
+            alternative_name: Alternative name to match. Defaults to None.
+            transform_possible_names: Functions to transform possible names.
+            threshold: Minimum similarity score, 0-100. Defaults to 65.0.
+
+        Returns:
+            Index of matching name from possible names or None
+        """
+        maxscore = None
+        matching_index = None
+        names = [name.lower()]
+        if alternative_name:
+            names.append(alternative_name.lower())
+        all_transforms = (lambda x: x, *transform_possible_names)
+        for i, possible_name in enumerate(possible_names):
+            for transform_possible_name in all_transforms:
+                transformed_possible_name = transform_possible_name(possible_name)
+                if not transformed_possible_name:
+                    continue
+                transformed_possible_name = transformed_possible_name.lower()
+                for query in names:
+                    score = (
+                        self.scorer(query, transformed_possible_name),
+                        rapidfuzz_fuzz.ratio(query, transformed_possible_name),
+                    )
+                    if maxscore is None or score > maxscore:
+                        maxscore = score
+                        matching_index = i
+        if maxscore is None or maxscore[0] < threshold:
+            return None
+        return matching_index
+
+
+def split_name(name: str, separators: re.Pattern) -> list[str]:
+    """
+    Split name on separators, dropping parts that are empty once normalised.
+
+    Args:
+        name: Name to split
+        separators: Compiled regular expression matching separators
+
+    Returns:
+        List of parts
+    """
+    parts = [part.strip() for part in separators.split(name)]
+    return [part for part in parts if normalise(part)]
+
+
+def resolve_name_parts(
+    name: str, name_to_code: dict[str, str], ignore: str | None = None
+) -> tuple[str | None, bool]:
+    """
+    Resolve a name made of parts split by separators (",", ";", "/", ":", "&",
+    newline, "and", "et") or, if there are none, by a hyphen. Parts are looked
+    up in name_to_code. A name with two parts gives a code if:
+    - one part's normalised form equals ignore and the other part has a code
+      (eg. "Falcón, Acosta" with the parent admin name "falcon")
+    - one part's code followed by "-" starts the other's (eg. "Protection -
+      GBV" gives PRO-GBV)
+    - both parts have the same code.
+    It is a list of names (eg. "Wash & Protection") if it has more than two
+    parts, or two parts with different codes (eg. "Guba-Khachmaz").
+
+    Args:
+        name: Name to resolve
+        name_to_code: Mapping from names (raw or normalised) to codes
+        ignore: Normalised name of a part that qualifies the other part.
+            Defaults to None.
+
+    Returns:
+        Tuple of code (or None) and whether name is a list of names
+    """
+    parts = split_name(name, LIST_SEPARATORS)
+    if len(parts) > 2:
+        return None, True
+    if len(parts) < 2:
+        parts = split_name(name, HYPHEN)
+        if len(parts) != 2:
+            return None, False
+    normalised_parts = [normalise(part) for part in parts]
+    if ignore in normalised_parts:
+        index = 1 - normalised_parts.index(ignore)
+        parts = [parts[index]]
+        normalised_parts = [normalised_parts[index]]
+    codes = []
+    for part, normalised_part in zip(parts, normalised_parts):
+        code = name_to_code.get(part, name_to_code.get(normalised_part))
+        if not code:
+            return None, False
+        codes.append(code)
+    code = codes[0]
+    if len(codes) == 1:
+        return code, False
+    other_code = codes[1]
+    if code == other_code or other_code.startswith(f"{code}-"):
+        return other_code, False
+    if code.startswith(f"{other_code}-"):
+        return code, False
+    return None, True
+
+
 def get_code_from_name(
     name: str,
     code_lookup: dict[str, str],
     unmatched: list[str],
     fuzzy_match: bool = True,
     match_threshold: int = 5,
+    matcher: Phonetics | RapidFuzzMatcher | None = None,
 ) -> str | None:
     """
-    Given a name (org type, sector, etc), return the corresponding code.
+    Given a name (org type, sector, etc), return the corresponding code. A
+    name made of parts is resolved by resolve_name_parts, and is not fuzzy
+    matched if it is a list of names.
 
     Args:
         name: Name to match
@@ -74,6 +275,7 @@ def get_code_from_name(
         unmatched: List of unmatched names
         fuzzy_match: Allow fuzzy matching or not
         match_threshold: Match threshold
+        matcher: Fuzzy matcher to use. Defaults to None (RapidFuzzMatcher).
 
     Returns:
         Matching code
@@ -91,11 +293,17 @@ def get_code_from_name(
     if len(name) <= match_threshold:
         unmatched.append(name)
         return None
-    if not fuzzy_match:
+    code, is_list = resolve_name_parts(name, code_lookup)
+    if code:
+        code_lookup[name] = code
+        return code
+    if not fuzzy_match or is_list:
         unmatched.append(name)
         return None
     names = [x for x in code_lookup.keys() if len(x) > match_threshold]
-    name_index = Phonetics().match(
+    if matcher is None:
+        matcher = RapidFuzzMatcher()
+    name_index = matcher.match(
         possible_names=names,
         name=name,
         alternative_name=name_clean,

{hdx_python_utilities-4.1.1 → hdx_python_utilities-4.2.0}/src/hdx/utilities/retriever.py

@@ -184,6 +184,8 @@ class Retrieve(BaseDownload):
         filename: str | None = None,
         logstr: str | None = None,
         fallback: bool = False,
+        resume: bool = False,
+        retries: int = 0,
         log_level: int = None,
         **kwargs: Any,
     ) -> Path:
@@ -194,6 +196,8 @@ class Retrieve(BaseDownload):
             filename: Filename of saved file. Defaults to getting from url.
             logstr: Text to use in log string to describe download. Defaults to filename.
             fallback: Whether to use static fallback if download fails. Defaults to False.
+            resume: Whether to resume a partial download using Range requests where the server supports it. Defaults to False.
+            retries: Number of times to retry a mid-stream failure. Only effective when resume=True. Defaults to 0.
             log_level: Level at which to log messages. Overrides level from constructor.
             **kwargs: Parameters to pass to download_file call
 
@@ -219,7 +223,9 @@ class Retrieve(BaseDownload):
                 log_level,
                 f"Downloading {logstr} from {self.get_url_logstr(url)} into {output_path}",
             )
-            return self.downloader.download_file(url, path=output_path, **kwargs)
+            return self.downloader.download_file(
+                url, path=output_path, resume=resume, retries=retries, **kwargs
+            )
         except DownloadError:
             if not fallback:
                 raise

hdx_python_utilities-4.2.0/tests/fixtures/downloader/resume_test.dat

@@ -0,0 +1,9 @@
+id,name,value
+1,alpha,10
+2,beta,20
+3,gamma,30
+4,delta,40
+5,epsilon,50
+6,zeta,60
+7,eta,70
+8,theta,80