Merge pull request #351 from mdekstrand/tweak/drop-deprecated-random

Remove legacy random code

conftest.py

@@ -1,59 +1,35 @@
 import logging
 from pytest import fixture
 
-import numpy as np
+from seedbank import numpy_rng, initialize
 
-from lenskit import util
+logging.getLogger("numba").setLevel(logging.INFO)
 
-logging.getLogger('numba').setLevel(logging.INFO)
-
-_log = logging.getLogger('lenskit.tests')
+_log = logging.getLogger("lenskit.tests")
 
 
 @fixture
 def rng():
-    return util.rng(42)
-
-
-@fixture
-def legacy_rng():
-    return util.rng(42, legacy_rng=True)
+    return numpy_rng(42)
 
 
 @fixture(autouse=True)
 def init_rng(request):
-    util.init_rng(42)
+    initialize(42)
 
 
 @fixture(autouse=True)
 def log_test(request):
-    modname = request.module.__name__ if request.module else '<unknown>'
-    funcname = request.function.__name__ if request.function else '<unknown>'
-    _log.info('running test %s:%s', modname, funcname)
-
-
-@fixture(autouse=True, scope='session')
-def carbon(request):
-    try:
-        from codecarbon import EmissionsTracker
-    except ImportError:
-        yield True  # we do nothing
-        return
-
-    tracker = EmissionsTracker("lkpy-tests", 5)
-    tracker.start()
-    try:
-        yield True
-    finally:
-        emissions = tracker.stop()
-        _log.info('test suite used %.3f kgCO2eq', emissions)
+    modname = request.module.__name__ if request.module else "<unknown>"
+    funcname = request.function.__name__ if request.function else "<unknown>"
+    _log.info("running test %s:%s", modname, funcname)
 
 
 def pytest_collection_modifyitems(items):
     # add 'slow' to all 'eval' tests
     for item in items:
-        evm = item.get_closest_marker('eval')
-        slm = item.get_closest_marker('slow')
+        evm = item.get_closest_marker("eval")
+        slm = item.get_closest_marker("slow")
         if evm is not None and slm is None:
-            _log.debug('adding slow mark to %s', item)
-            item.add_marker('slow')
+            _log.debug("adding slow mark to %s", item)
+            item.add_marker("slow")

lenskit/algorithms/als.py

@@ -1,11 +1,11 @@
 import logging
 from collections import namedtuple
-import warnings
 
 import numpy as np
 from numba import njit, prange
 
 from csr import CSR
+from seedbank import numpy_rng
 
 from .bias import Bias
 from .mf_common import MFPredictor
@@ -337,7 +337,7 @@ class BiasedMF(MFPredictor):
             damping ``damping``.
         method(str): the solver to use (see above).
         rng_spec:
-            Random number generator or state (see :func:`lenskit.util.random.rng`).
+            Random number generator or state (see :func:`seedbank.numpy_rng`).
         progress: a :func:`tqdm.tqdm`-compatible progress bar function
     """
 
@@ -366,7 +366,7 @@ def __init__(
         else:
             self.bias = bias
         self.progress = progress if progress is not None else util.no_progress
-        self.rng = util.rng(rng_spec)
+        self.rng = numpy_rng(rng_spec)
         self.save_user_features = save_user_features
 
     def fit(self, ratings, **kwargs):
@@ -600,7 +600,7 @@ def __init__(
         self.weight = weight
         self.use_ratings = use_ratings
         self.method = method
-        self.rng = util.rng(rng_spec)
+        self.rng = numpy_rng(rng_spec)
         self.progress = progress if progress is not None else util.no_progress
         self.save_user_features = save_user_features
 

lenskit/algorithms/basic.py

@@ -287,7 +287,7 @@ def __init__(self, selector=None, rng_spec=None):
         else:
             self.selector = UnratedItemCandidateSelector()
         # Get a Pandas-compatible RNG
-        self.rng_source = derivable_rng(rng_spec, legacy=True)
+        self.rng_source = derivable_rng(rng_spec)
         self.items = None
 
     def fit(self, ratings, **kwargs):

lenskit/algorithms/funksvd.py

@@ -8,7 +8,7 @@
 import pandas as pd
 import numpy as np
 import numba as n
-from pandas.core.series import Series
+from seedbank import numpy_rng
 
 try:
     from numba.experimental import jitclass
@@ -242,7 +242,7 @@ def __init__(
             self.bias = Bias(damping=damping)
         else:
             self.bias = bias
-        self.random = util.rng(random_state)
+        self.random = numpy_rng(random_state)
 
     def fit(self, ratings, **kwargs):
         """

lenskit/crossfold.py

@@ -8,7 +8,7 @@
 
 import numpy as np
 import pandas as pd
-from . import util
+from seedbank import numpy_rng
 
 TTPair = namedtuple("TTPair", ["train", "test"])
 TTPair.__doc__ = "Train-test pair (named tuple)."
@@ -29,7 +29,7 @@ def partition_rows(data, partitions, *, rng_spec=None):
         partitions(int):
             The number of partitions to produce.
         rng_spec:
-            The random number generator or seed (see :py:func:`lenskit.util.rng`).
+            The random number generator or seed (see :py:func:`seedbank.numpy_rng`).
 
     Returns:
         iterator: an iterator of train-test pairs
@@ -41,7 +41,7 @@ def partition_rows(data, partitions, *, rng_spec=None):
     # create an array of indexes
     rows = np.arange(len(data))
     # shuffle the indices & split into partitions
-    rng = util.rng(rng_spec)
+    rng = numpy_rng(rng_spec)
     rng.shuffle(rows)
     test_sets = np.array_split(rows, partitions)
 
@@ -90,15 +90,16 @@ def sample_rows(data, partitions, size, disjoint=True, *, rng_spec=None):
         disjoint(bool):
             If ``True``, force samples to be disjoint.
         rng_spec:
-            The random number generator or seed (see :py:func:`lenskit.util.rng`).
+            The random number generator or seed (see :py:func:`seedbank.numpy_rng`).
 
     Returns:
         iterator: An iterator of train-test pairs.
     """
 
     confirm_unique_index(data)
+    rng = numpy_rng(rng_spec)
     if partitions is None:
-        test = data.sample(n=size)
+        test = data.sample(n=size, random_state=rng)
         tr_mask = pd.Series(True, index=data.index)
         tr_mask.loc[test.index] = False
         train = data[tr_mask]
@@ -111,13 +112,11 @@ def sample_rows(data, partitions, size, disjoint=True, *, rng_spec=None):
             size,
             len(data),
         )
-        return partition_rows(data, partitions)
+        return partition_rows(data, partitions, rng_spec=rng)
 
     # create an array of indexes
     rows = np.arange(len(data))
 
-    rng = util.rng(rng_spec)
-
     if disjoint:
         _logger.info("creating %d disjoint samples of size %d", partitions, size)
         ips = _disjoint_sample(rows, partitions, size, rng)
@@ -181,7 +180,7 @@ class SampleN(PartitionMethod):
 
     def __init__(self, n, rng_spec=None):
         self.n = n
-        self.rng = util.rng(rng_spec, legacy=True)
+        self.rng = numpy_rng(rng_spec)
 
     def __call__(self, udf):
         return udf.sample(n=self.n, random_state=self.rng)
@@ -197,7 +196,7 @@ class SampleFrac(PartitionMethod):
 
     def __init__(self, frac, rng_spec=None):
         self.fraction = frac
-        self.rng = util.rng(rng_spec, legacy=True)
+        self.rng = numpy_rng(rng_spec)
 
     def __call__(self, udf):
         return udf.sample(frac=self.fraction, random_state=self.rng)
@@ -247,7 +246,7 @@ def partition_users(data, partitions: int, method: PartitionMethod, *, rng_spec=
         data(pandas.DataFrame): a data frame containing ratings or other data you wish to partition.
         partitions(int): the number of partitions to produce
         method(PartitionMethod): The method for selecting test rows for each user.
-        rng_spec: The random number generator or seed (see :py:func:`lenskit.util.rng`).
+        rng_spec: The random number generator or seed (see :py:func:`seedbank.numpy_rng`).
 
     Returns
         iterator: an iterator of train-test pairs
@@ -263,7 +262,7 @@ def partition_users(data, partitions: int, method: PartitionMethod, *, rng_spec=
     # create an array of indexes into user row
     rows = np.arange(len(users))
     # shuffle the indices & split into partitions
-    rng = util.rng(rng_spec, legacy=True)
+    rng = numpy_rng(rng_spec)
     rng.shuffle(rows)
     test_sets = np.array_split(rows, partitions)
 
@@ -303,14 +302,14 @@ def sample_users(
         method(PartitionMethod):
             The method for obtaining user test ratings.
         rng_spec:
-            The random number generator or seed (see :py:func:`lenskit.util.rng`).
+            The random number generator or seed (see :py:func:`seedbank.numpy_rng`).
 
     Returns:
         iterator: An iterator of train-test pairs (as :class:`TTPair` objects).
     """
 
     confirm_unique_index(data)
-    rng = util.rng(rng_spec, legacy=True)
+    rng = numpy_rng(rng_spec)
 
     user_col = data["user"]
     users = user_col.unique()

lenskit/util/__init__.py

@@ -9,7 +9,7 @@
 from ..algorithms import Algorithm
 from .log import log_to_notebook, log_to_stderr  # noqa: F401
 from .timing import Stopwatch  # noqa: F401
-from .random import rng, init_rng, derivable_rng  # noqa: F401
+from .random import derivable_rng
 from .parallel import proc_count  # noqa: F401
 
 try:
@@ -24,8 +24,6 @@
     "log_to_notebook",
     "Stopwatch",
     "read_df_detect",
-    "rng",
-    "init_rng",
     "derivable_rng",
     "proc_count",
     "clone",

lenskit/util/parallel.py

@@ -14,9 +14,10 @@
 import pickle
 from threadpoolctl import threadpool_limits
 
+import seedbank
+
 from lenskit.sharing import persist, PersistedModel
 from lenskit.util.log import log_queue
-from lenskit.util.random import derive_seed, init_rng, get_root_seed
 
 _log = logging.getLogger(__name__)
 __work_model = None
@@ -91,7 +92,7 @@ def _initialize_worker(log_queue, seed):
     __is_worker = True
     faulthandler.enable()
     if seed is not None:
-        init_rng(seed)
+        seedbank.initialize(seed)
     if log_queue is not None:
         h = logging.handlers.QueueHandler(log_queue)
         root = logging.getLogger()
@@ -101,7 +102,7 @@ def _initialize_worker(log_queue, seed):
 
 
 def _initialize_mp_worker(mkey, func, threads, log_queue, seed):
-    seed = derive_seed(mp.current_process().name, base=seed)
+    seed = seedbank.derive_seed(mp.current_process().name, base=seed)
     _initialize_worker(log_queue, seed)
     global __work_model, __work_func
 
@@ -183,7 +184,7 @@ def run_sp(func, *args, **kwargs):
     """
     ctx = LKContext.INSTANCE
     rq = ctx.SimpleQueue()
-    seed = derive_seed()
+    seed = seedbank.derive_seed()
     worker_args = (log_queue(), seed, rq, func, args, kwargs)
     _log.debug("spawning subprocess to run %s", func)
     proc = ctx.Process(target=_sp_worker, args=worker_args)
@@ -300,7 +301,10 @@ def __init__(self, model, func, n_jobs, persist_method):
         os.environ["_LK_IN_MP"] = "yes"
         kid_tc = proc_count(level=1)
         self.executor = ProcessPoolExecutor(
-            n_jobs, ctx, _initialize_mp_worker, (key, func, kid_tc, log_queue(), get_root_seed())
+            n_jobs,
+            ctx,
+            _initialize_mp_worker,
+            (key, func, kid_tc, log_queue(), seedbank.root_seed()),
         )
 
     def map(self, *iterables):