replay-rec 0.21.0rc0__tar.gz → 0.21.1__tar.gz

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: replay-rec
-Version: 0.21.0rc0
+Version: 0.21.1
 Summary: RecSys Library
 License-Expression: Apache-2.0
 License-File: LICENSE
@@ -14,23 +14,29 @@ Classifier: Intended Audience :: Developers
 Classifier: Intended Audience :: Science/Research
 Classifier: Natural Language :: English
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
-Requires-Dist: d3rlpy (>=2.8.1,<2.9)
-Requires-Dist: implicit (>=0.7.2,<0.8)
-Requires-Dist: lightautoml (>=0.4.1,<0.5)
-Requires-Dist: lightning (>=2.0.2,<=2.4.0)
-Requires-Dist: numba (>=0.50,<1)
+Provides-Extra: spark
+Provides-Extra: torch
+Provides-Extra: torch-cpu
+Requires-Dist: lightning (<2.6.0) ; extra == "torch" or extra == "torch-cpu"
+Requires-Dist: lightning ; extra == "torch"
+Requires-Dist: lightning ; extra == "torch-cpu"
 Requires-Dist: numpy (>=1.20.0,<2)
 Requires-Dist: pandas (>=1.3.5,<2.4.0)
 Requires-Dist: polars (<2.0)
-Requires-Dist: psutil (<=7.0.0)
+Requires-Dist: psutil (<=7.0.0) ; extra == "spark"
+Requires-Dist: psutil ; extra == "spark"
 Requires-Dist: pyarrow (<22.0)
-Requires-Dist: pyspark (>=3.0,<3.5)
-Requires-Dist: pytorch-optimizer (>=3.8.0,<4)
-Requires-Dist: sb-obp (>=0.5.10,<0.6)
+Requires-Dist: pyspark (>=3.0,<3.5) ; extra == "spark"
+Requires-Dist: pyspark ; extra == "spark"
+Requires-Dist: pytorch-optimizer (>=3.8.0,<3.9.0) ; extra == "torch" or extra == "torch-cpu"
+Requires-Dist: pytorch-optimizer ; extra == "torch"
+Requires-Dist: pytorch-optimizer ; extra == "torch-cpu"
 Requires-Dist: scikit-learn (>=1.6.1,<1.7.0)
 Requires-Dist: scipy (>=1.8.1,<2.0.0)
 Requires-Dist: setuptools
-Requires-Dist: torch (>=1.8,<3.0.0)
+Requires-Dist: torch (>=1.8,<3.0.0) ; extra == "torch" or extra == "torch-cpu"
+Requires-Dist: torch ; extra == "torch"
+Requires-Dist: torch ; extra == "torch-cpu"
 Requires-Dist: tqdm (>=4.67,<5)
 Project-URL: Homepage, https://sb-ai-lab.github.io/RePlay/
 Project-URL: Repository, https://github.com/sb-ai-lab/RePlay

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/pyproject.toml

@@ -65,19 +65,19 @@ dependencies = [
     "scikit-learn (>=1.6.1,<1.7.0)",
     "pyarrow (<22.0)",
     "tqdm (>=4.67,<5)",
-    "torch (>=1.8,<3.0.0)",
-    "lightning (>=2.0.2,<=2.4.0)",
-    "pytorch-optimizer (>=3.8.0,<4)",
-    "lightautoml (>=0.4.1,<0.5)",
-    "numba (>=0.50,<1)",
-    "sb-obp (>=0.5.10,<0.6)",
-    "d3rlpy (>=2.8.1,<2.9)",
-    "implicit (>=0.7.2,<0.8)",
-    "pyspark (>=3.0,<3.5)",
-    "psutil (<=7.0.0)",
+    "pyspark (>=3.0,<3.5); extra == 'spark'",
+    "psutil (<=7.0.0); extra == 'spark'",
+    "torch (>=1.8, <3.0.0); extra == 'torch' or extra == 'torch-cpu'",
+    "pytorch-optimizer (>=3.8.0,<3.9.0); extra == 'torch' or extra == 'torch-cpu'",
+    "lightning (<2.6.0); extra == 'torch' or extra == 'torch-cpu'",
 ]
 dynamic = ["dependencies"]
-version = "0.21.0.preview"
+version = "0.21.1"
+
+[project.optional-dependencies]
+spark = ["pyspark", "psutil"]
+torch = ["torch", "pytorch-optimizer", "lightning"]
+torch-cpu = ["torch", "pytorch-optimizer", "lightning"]
 
 [project.urls]
 homepage = "https://sb-ai-lab.github.io/RePlay/"
@@ -91,6 +91,13 @@ target-version = ["py39", "py310", "py311", "py312"]
 packages = [{include = "replay"}]
 exclude = [
     "replay/conftest.py",
+    "replay/experimental",
+]
+
+[tool.poetry.dependencies]
+torch = [
+    {markers = "extra == 'torch-cpu' and extra !='torch'", source = "torch-cpu-mirror"},
+    {markers = "extra == 'torch' and extra !='torch-cpu'", source = "PyPI"},
 ]
 
 [[tool.poetry.source]]
@@ -100,7 +107,7 @@ priority = "explicit"
 
 [tool.poetry-dynamic-versioning]
 enable = false
-format-jinja = """0.21.0{{ env['PACKAGE_SUFFIX'] }}"""
+format-jinja = """0.21.1{{ env['PACKAGE_SUFFIX'] }}"""
 vcs = "git"
 
 [tool.ruff]

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/__init__.py

@@ -4,4 +4,4 @@
 # functionality removed in Python 3.12 is used in downstream packages (like lightfm)
 import setuptools as _
 
-__version__ = "0.21.0.preview"
+__version__ = "0.21.1"

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/data/nn/parquet/parquet_module.py

@@ -94,7 +94,7 @@ class ParquetModule(L.LightningDataModule):
         missing_splits = [split_name for split_name, split_path in self.datapaths.items() if split_path is None]
         if missing_splits:
             msg = (
-                f"The following dataset paths aren't provided: {','.join(missing_splits)}."
+                f"The following dataset paths aren't provided: {','.join(missing_splits)}. "
                 "Make sure to disable these stages in your Lightning Trainer configuration."
             )
             warnings.warn(msg, stacklevel=2)

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/metrics/torch_metrics_builder.py

@@ -400,7 +400,7 @@ def metrics_to_df(metrics: Mapping[str, float]) -> PandasDataFrame:
 
     metric_name_and_k = metrics_df["m"].str.split("@", expand=True)
     metrics_df["metric"] = metric_name_and_k[0]
-    metrics_df["k"] = metric_name_and_k[1]
+    metrics_df["k"] = metric_name_and_k[1].astype(int)
 
     pivoted_metrics = metrics_df.pivot(index="metric", columns="k", values="v")
     pivoted_metrics.index.name = None

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/models/nn/sequential/callbacks/validation_callback.py

@@ -162,14 +162,24 @@ class ValidationMetricsCallback(lightning.Callback):
         @rank_zero_only
         def print_metrics() -> None:
             metrics = {}
+
             for name, value in trainer.logged_metrics.items():
                 if "@" in name:
                     metrics[name] = value.item()
 
-            if metrics:
-                metrics_df = metrics_to_df(metrics)
+            if not metrics:
+                return
 
-                print(metrics_df)  # noqa: T201
-                print()  # noqa: T201
+            if len(self._dataloaders_size) > 1:
+                for i in range(len(self._dataloaders_size)):
+                    suffix = trainer._results.DATALOADER_SUFFIX.format(i)[1:]
+                    cur_dataloader_metrics = {k.split("/")[0]: v for k, v in metrics.items() if suffix in k}
+                    metrics_df = metrics_to_df(cur_dataloader_metrics)
+
+                    print(suffix)  # noqa: T201
+                    print(metrics_df, "\n")  # noqa: T201
+            else:
+                metrics_df = metrics_to_df(metrics)
+                print(metrics_df, "\n")  # noqa: T201
 
         print_metrics()

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/nn/lightning/callback/metrics_callback.py

@@ -2,7 +2,6 @@ from typing import Any, Optional
 
 import lightning
 import torch
-from lightning.pytorch.utilities.combined_loader import CombinedLoader
 from lightning.pytorch.utilities.rank_zero import rank_zero_only
 
 from replay.metrics.torch_metrics_builder import (
@@ -64,8 +63,8 @@ class ComputeMetricsCallback(lightning.Callback):
         self._train_column = train_column
 
     def _get_dataloaders_size(self, dataloaders: Optional[Any]) -> list[int]:
-        if isinstance(dataloaders, CombinedLoader):
-            return [len(dataloader) for dataloader in dataloaders.flattened]  # pragma: no cover
+        if isinstance(dataloaders, list):
+            return [len(dataloader) for dataloader in dataloaders]
         return [len(dataloaders)]
 
     def on_validation_epoch_start(
@@ -123,7 +122,7 @@ class ComputeMetricsCallback(lightning.Callback):
         batch: dict,
         batch_idx: int,
         dataloader_idx: int = 0,
-    ) -> None:  # pragma: no cover
+    ) -> None:
         self._batch_end(
             trainer,
             pl_module,
@@ -159,7 +158,7 @@ class ComputeMetricsCallback(lightning.Callback):
     def on_validation_epoch_end(self, trainer: lightning.Trainer, pl_module: LightningModule) -> None:
         self._epoch_end(trainer, pl_module)
 
-    def on_test_epoch_end(self, trainer: lightning.Trainer, pl_module: LightningModule) -> None:  # pragma: no cover
+    def on_test_epoch_end(self, trainer: lightning.Trainer, pl_module: LightningModule) -> None:
         self._epoch_end(trainer, pl_module)
 
     def _epoch_end(
@@ -170,14 +169,24 @@ class ComputeMetricsCallback(lightning.Callback):
         @rank_zero_only
         def print_metrics() -> None:
             metrics = {}
+
             for name, value in trainer.logged_metrics.items():
                 if "@" in name:
                     metrics[name] = value.item()
 
-            if metrics:
-                metrics_df = metrics_to_df(metrics)
+            if not metrics:
+                return
 
-                print(metrics_df)  # noqa: T201
-                print()  # noqa: T201
+            if len(self._dataloaders_size) > 1:
+                for i in range(len(self._dataloaders_size)):
+                    suffix = trainer._results.DATALOADER_SUFFIX.format(i)[1:]
+                    cur_dataloader_metrics = {k.split("/")[0]: v for k, v in metrics.items() if suffix in k}
+                    metrics_df = metrics_to_df(cur_dataloader_metrics)
+
+                    print(suffix)  # noqa: T201
+                    print(metrics_df, "\n")  # noqa: T201
+            else:
+                metrics_df = metrics_to_df(metrics)
+                print(metrics_df, "\n")  # noqa: T201
 
         print_metrics()

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/nn/lightning/callback/predictions_callback.py

@@ -15,11 +15,11 @@ from replay.utils import (
     SparkDataFrame,
 )
 
-if PYSPARK_AVAILABLE:  # pragma: no cover
+if PYSPARK_AVAILABLE:
     import pyspark.sql.functions as sf
     from pyspark.sql import SparkSession
     from pyspark.sql.types import ArrayType, DoubleType, IntegerType, StructType
-else:  # pragma: no cover
+else:
     SparkSession = MissingImport
 
 

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/nn/loss/base.py

@@ -85,7 +85,7 @@ class SampledLossBase(torch.nn.Module):
             # [batch_size, num_negatives] -> [batch_size, 1, num_negatives]
             negative_labels = negative_labels.unsqueeze(1).repeat(1, seq_len, 1)
 
-        if negative_labels.dim() == 3:  # pragma: no cover
+        if negative_labels.dim() == 3:
             # [batch_size, seq_len, num_negatives] -> [batch_size, seq_len, 1, num_negatives]
             negative_labels = negative_labels.unsqueeze(-2)
             if num_positives != 1:
@@ -119,7 +119,7 @@ class SampledLossBase(torch.nn.Module):
         positive_labels = positive_labels[target_padding_mask].unsqueeze(-1)
         assert positive_labels.size() == (masked_batch_size, 1)
 
-        if negative_labels.dim() != 1:  # pragma: no cover
+        if negative_labels.dim() != 1:
             # [batch_size, seq_len, num_positives, num_negatives] -> [masked_batch_size, num_negatives]
             negative_labels = negative_labels[target_padding_mask]
             assert negative_labels.size() == (masked_batch_size, num_negatives)
@@ -183,7 +183,7 @@ def mask_negative_logits(
     if negative_labels_ignore_index >= 0:
         negative_logits.masked_fill_(negative_labels == negative_labels_ignore_index, -1e9)
 
-    if negative_labels.dim() > 1:  # pragma: no cover
+    if negative_labels.dim() > 1:
         # [masked_batch_size, num_negatives] -> [masked_batch_size, 1, num_negatives]
         negative_labels = negative_labels.unsqueeze(-2)
 

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/nn/loss/login_ce.py

@@ -74,7 +74,7 @@ class LogInCEBase(SampledLossBase):
         positive_labels = positive_labels[masked_target_padding_mask]
         assert positive_labels.size() == (masked_batch_size, num_positives)
 
-        if negative_labels.dim() > 1:  # pragma: no cover
+        if negative_labels.dim() > 1:
             # [batch_size, seq_len, num_negatives] -> [masked_batch_size, num_negatives]
             negative_labels = negative_labels[masked_target_padding_mask]
             assert negative_labels.size() == (masked_batch_size, num_negatives)

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/nn/sequential/sasrec/model.py

@@ -141,7 +141,7 @@ class SasRec(torch.nn.Module):
                     feature_type=FeatureType.CATEGORICAL,
                     embedding_dim=256,
                     padding_value=NUM_UNIQUE_ITEMS,
-                    cardinality=NUM_UNIQUE_ITEMS+1,
+                    cardinality=NUM_UNIQUE_ITEMS,
                     feature_hint=FeatureHint.ITEM_ID,
                     feature_sources=[TensorFeatureSource(FeatureSource.INTERACTIONS, "item_id")]
                 ),

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/nn/sequential/twotower/reader.py

@@ -22,7 +22,6 @@ class FeaturesReader:
         :param schema: the same tensor schema used in TwoTower model.
         :param metadata: A dictionary of feature names that
             associated with its shape and padding_value.\n
-            Example: {"item_id" : {"shape": 100, "padding": 7657}}.\n
             For details, see the section :ref:`parquet-processing`.
         :param path: path to parquet with dataframe of item features.\n
             **Note:**\n
@@ -30,8 +29,8 @@ class FeaturesReader:
             2. Every feature for item "tower" in `schema` must contain ``feature_sources`` with the names
                of the source features to create correct inverse mapping.
                Also, for each such feature one of the requirements must be met: the ``schema`` for the feature must
-               contain ``feature_sources`` with a source of type FeatureSource.ITEM_FEATURES
-               or hint type FeatureHint.ITEM_ID.
+               contain ``feature_sources`` with a source of type ``FeatureSource.ITEM_FEATURES``
+               or hint type ``FeatureHint.ITEM_ID``.
 
         """
         item_feature_names = [
@@ -81,8 +80,18 @@ class FeaturesReader:
         self._features = {}
 
         for k in features.columns:
-            dtype = torch.float32 if schema[k].is_num else torch.int64
-            feature_tensor = torch.asarray(features[k], dtype=dtype)
+            dtype = np.float32 if schema[k].is_num else np.int64
+            if schema[k].is_list:
+                feature = np.asarray(
+                    features[k].to_list(),
+                    dtype=dtype,
+                )
+            else:
+                feature = features[k].to_numpy(dtype=dtype)
+            feature_tensor = torch.asarray(
+                feature,
+                dtype=torch.float32 if schema[k].is_num else torch.int64,
+            )
             self._features[k] = feature_tensor
 
     def __getitem__(self, key: str) -> torch.Tensor:

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/nn/transform/template/sasrec.py

@@ -14,7 +14,7 @@ def make_default_sasrec_transforms(
 
     Generated pipeline expects input dataset to contain the following columns:
         1) Query ID column, specified by ``query_column``.
-        2) Item ID column, specified in the tensor schema.
+        2) All features specified in the ``tensor_schema``.
 
     :param tensor_schema: TensorSchema used to infer feature columns.
     :param query_column: Name of the column containing query IDs. Default: ``"query_id"``.
@@ -32,12 +32,12 @@ def make_default_sasrec_transforms(
         ),
         UnsqueezeTransform("target_padding_mask", -1),
         UnsqueezeTransform("positive_labels", -1),
-        GroupTransform({"feature_tensors": [item_column]}),
+        GroupTransform({"feature_tensors": tensor_schema.names}),
     ]
 
     val_transforms = [
         RenameTransform({query_column: "query_id", f"{item_column}_mask": "padding_mask"}),
-        GroupTransform({"feature_tensors": [item_column]}),
+        GroupTransform({"feature_tensors": tensor_schema.names}),
     ]
     test_transforms = copy.deepcopy(val_transforms)
 

{replay_rec-0.21.0rc0 → replay_rec-0.21.1}/replay/nn/transform/template/twotower.py

@@ -13,7 +13,7 @@ def make_default_twotower_transforms(
 
     Generated pipeline expects input dataset to contain the following columns:
         1) Query ID column, specified by ``query_column``.
-        2) Item ID column, specified in the tensor schema.
+        2) All features specified in the ``tensor_schema``.
 
     :param tensor_schema: TensorSchema used to infer feature columns.
     :param query_column: Name of the column containing query IDs. Default: ``"query_id"``.

replay_rec-0.21.0rc0/replay/experimental/metrics/__init__.py

@@ -1,62 +0,0 @@
-"""
-Most metrics require dataframe with recommendations
-and dataframe with ground truth values —
-which objects each user interacted with.
-
-- recommendations (Union[pandas.DataFrame, spark.DataFrame]):
-    predictions of a recommender system,
-    DataFrame with columns ``[user_id, item_id, relevance]``
-- ground_truth (Union[pandas.DataFrame, spark.DataFrame]):
-    test data, DataFrame with columns
-    ``[user_id, item_id, timestamp, relevance]``
-
-Metric is calculated for all users, presented in ``ground_truth``
-for accurate metric calculation in case when the recommender system generated
-recommendation not for all users.  It is assumed, that all users,
-we want to calculate metric for, have positive interactions.
-
-But if we have users, who observed the recommendations, but have not responded,
-those users will be ignored and metric will be overestimated.
-For such case we propose additional optional parameter ``ground_truth_users``,
-the dataframe with all users, which should be considered during the metric calculation.
-
-- ground_truth_users (Optional[Union[pandas.DataFrame, spark.DataFrame]]):
-    full list of users to calculate metric for, DataFrame with ``user_id`` column
-
-Every metric is calculated using top ``K`` items for each user.
-It is also possible to calculate metrics
-using multiple values for ``K`` simultaneously.
-In this case the result will be a dictionary and not a number.
-
-Make sure your recommendations do not contain user-item duplicates
-as duplicates could lead to the wrong calculation results.
-
-- k (Union[Iterable[int], int]):
-    a single number or a list, specifying the
-    truncation length for recommendation list for each user
-
-By default, metrics are averaged by users,
-but you can alternatively use method ``metric.median``.
-Also, you can get the lower bound
-of ``conf_interval`` for a given ``alpha``.
-
-Diversity metrics require extra parameters on initialization stage,
-but do not use ``ground_truth`` parameter.
-
-For each metric, a formula for its calculation is given, because this is
-important for the correct comparison of algorithms, as mentioned in our
-`article <https://arxiv.org/abs/2206.12858>`_.
-"""
-
-from replay.experimental.metrics.base_metric import Metric, NCISMetric
-from replay.experimental.metrics.coverage import Coverage
-from replay.experimental.metrics.hitrate import HitRate
-from replay.experimental.metrics.map import MAP
-from replay.experimental.metrics.mrr import MRR
-from replay.experimental.metrics.ncis_precision import NCISPrecision
-from replay.experimental.metrics.ndcg import NDCG
-from replay.experimental.metrics.precision import Precision
-from replay.experimental.metrics.recall import Recall
-from replay.experimental.metrics.rocauc import RocAuc
-from replay.experimental.metrics.surprisal import Surprisal
-from replay.experimental.metrics.unexpectedness import Unexpectedness