Project import generated by Copybara. (#20)

bazel/requirements/BUILD.bazel

@@ -21,12 +21,14 @@ _GENERATE_TOOL = ":parse_and_generate_requirements"
 
 _GENERATE_COMMAND = "$(location " + _GENERATE_TOOL + ") $(location " + _SRC_REQUIREMENT_FILE + ") --schema $(location " + _SCHEMA_FILE + ") {options} > $@"
 
-_TEMPLATE_FOLDER_PATH = "//bazel/requirements/templates"
-
 _AUTOGEN_HEADERS = """# DO NOT EDIT!
 # Generate by running 'bazel run //bazel/requirements:sync_requirements'
 """
 
+# "---" is a document start marker, which is legit but optional (https://yaml.org/spec/1.1/#c-document-start). This
+# is needed for conda meta.yaml to bypass some bug from conda side.
+_YAML_START_DOCUMENT_MARKER = "---"
+
 _GENERATED_REQUIREMENTS_FILES = {
     "requirements_txt": {
         "cmd": "--mode dev_version --format text",
@@ -77,7 +79,7 @@ _GENERATED_REQUIREMENTS_FILES = {
             "{generated}.body".format(generated = value["generated"]),
         ],
         outs = [value["generated"]],
-        cmd = "(echo -e \""+ _AUTOGEN_HEADERS +"\" ; cat $(location :{generated}.body) ) > $@".format(
+        cmd = "(echo -e \"" + _AUTOGEN_HEADERS + "\" ; cat $(location :{generated}.body) ) > $@".format(
             generated = value["generated"],
         ),
         tools = [_GENERATE_TOOL],
@@ -99,15 +101,24 @@ genrule(
 )
 
 yq(
-    name = "gen_conda_meta",
+    name = "gen_conda_meta_body_format",
     srcs = [
         ":meta.body.yaml",
-        "{template_folder}:meta.tpl.yaml".format(template_folder = _TEMPLATE_FOLDER_PATH),
+        "//bazel/requirements/templates:meta.tpl.yaml",
     ],
-    outs = ["meta.yaml"],
+    outs = ["meta.body.formatted.yaml"],
     expression = ". as $item ireduce ({}; . * $item ) | sort_keys(..)",
 )
 
+genrule(
+    name = "gen_conda_meta",
+    srcs = [
+        ":meta.body.formatted.yaml",
+    ],
+    outs = ["meta.yaml"],
+    cmd = "(echo -e \"" + _AUTOGEN_HEADERS + "\" ; echo \"" + _YAML_START_DOCUMENT_MARKER + "\"; cat $(location :meta.body.formatted.yaml) ) > $@",
+)
+
 # Create a test target for each file that Bazel should
 # write to the source tree.
 [

bazel/requirements/templates/meta.tpl.yaml

@@ -1,9 +1,3 @@
-# DO NOT EDIT!
-# Generated by //bazel/requirements:gen_conda_meta
-# To update, run:
-# bazel run //bazel/requirements:sync_requirements
-#
-
 package:
   name: snowflake-ml-python
 

ci/conda_recipe/meta.yaml

@@ -1,8 +1,7 @@
 # DO NOT EDIT!
-# Generated by //bazel/requirements:gen_conda_meta
-# To update, run:
-# bazel run //bazel/requirements:sync_requirements
-#
+# Generate by running 'bazel run //bazel/requirements:sync_requirements'
+
+---
 about:
   description: |
     Snowflake ML client Library is used for interacting with Snowflake to build machine learning solutions.

ci/get_excluded_tests.sh

@@ -12,7 +12,7 @@
 #     The missing dependency cuold happen when a new operator is being developed, but not yet released.
 
 set -o pipefail
-set -eu
+set -u
 
 echo "Running "$0
 

snowflake/ml/modeling/impute/simple_imputer.py

@@ -15,6 +15,8 @@
 from snowflake.snowpark import functions as F, types as T
 from snowflake.snowpark._internal import utils as snowpark_utils
 
+_SUBPROJECT = "Impute"
+
 STRATEGY_TO_STATE_DICT = {
     "constant": None,
     "mean": _utils.NumericStatistics.MEAN,
@@ -194,10 +196,7 @@ def check_type_consistency(col_types: Dict[str, T.DataType]) -> None:
 
         return input_col_datatypes
 
-    @telemetry.send_api_usage_telemetry(
-        project=base.PROJECT,
-        subproject=base.SUBPROJECT,
-    )
+    @telemetry.send_api_usage_telemetry(project=base.PROJECT, subproject=_SUBPROJECT)
     def fit(self, dataset: snowpark.DataFrame) -> "SimpleImputer":
         """
         Compute values to impute for the dataset according to the strategy.
@@ -214,7 +213,7 @@ def fit(self, dataset: snowpark.DataFrame) -> "SimpleImputer":
         input_col_datatypes = self._get_dataset_input_col_datatypes(dataset)
 
         self.statistics_: Dict[str, Any] = {}
-        statement_params = telemetry.get_statement_params(base.PROJECT, base.SUBPROJECT, self.__class__.__name__)
+        statement_params = telemetry.get_statement_params(base.PROJECT, _SUBPROJECT, self.__class__.__name__)
 
         if self.strategy == "constant":
             if self.fill_value is None:
@@ -274,14 +273,8 @@ def fit(self, dataset: snowpark.DataFrame) -> "SimpleImputer":
         self._is_fitted = True
         return self
 
-    @telemetry.send_api_usage_telemetry(
-        project=base.PROJECT,
-        subproject=base.SUBPROJECT,
-    )
-    @telemetry.add_stmt_params_to_df(
-        project=base.PROJECT,
-        subproject=base.SUBPROJECT,
-    )
+    @telemetry.send_api_usage_telemetry(project=base.PROJECT, subproject=_SUBPROJECT)
+    @telemetry.add_stmt_params_to_df(project=base.PROJECT, subproject=_SUBPROJECT)
     def transform(self, dataset: Union[snowpark.DataFrame, pd.DataFrame]) -> Union[snowpark.DataFrame, pd.DataFrame]:
         """
         Transform the input dataset by imputing the computed statistics in the input columns.

snowflake/ml/modeling/metrics/BUILD.bazel

@@ -13,6 +13,7 @@ py_library(
         "precision_recall_fscore_support.py",
         "precision_score.py",
         "regression.py",
+        "roc_curve.py",
     ],
     deps = [
         ":init",

snowflake/ml/modeling/metrics/__init__.py

@@ -4,6 +4,7 @@
 from .covariance import covariance
 from .precision_recall_fscore_support import precision_recall_fscore_support
 from .precision_score import precision_score
+from .roc_curve import roc_curve
 
 __all__ = [
     "accuracy_score",
@@ -12,4 +13,5 @@
     "covariance",
     "precision_recall_fscore_support",
     "precision_score",
+    "roc_curve",
 ]

snowflake/ml/modeling/metrics/precision_recall_fscore_support.py

@@ -115,10 +115,18 @@ def precision_recall_fscore_support(
 
     session = df._session
     assert session is not None
-    query = df.queries["queries"][-1]
     sproc_name = f"precision_recall_fscore_support_{snowpark_utils.generate_random_alphanumeric()}"
     statement_params = telemetry.get_statement_params(_PROJECT, _SUBPROJECT)
 
+    cols = []
+    if isinstance(y_true_col_names, str):
+        cols = [y_true_col_names, y_pred_col_names]
+    elif isinstance(y_true_col_names, list):
+        cols = y_true_col_names + y_pred_col_names  # type:ignore[assignment, operator]
+    if sample_weight_col_name:
+        cols.append(sample_weight_col_name)
+    query = df[cols].queries["queries"][-1]
+
     @F.sproc(  # type: ignore[misc]
         session=session,
         name=sproc_name,

snowflake/ml/modeling/metrics/roc_curve.py

@@ -0,0 +1,94 @@
+from typing import Optional, Tuple, Union
+
+import cloudpickle
+import numpy.typing as npt
+from sklearn import metrics
+
+from snowflake import snowpark
+from snowflake.ml._internal import telemetry
+from snowflake.snowpark import functions as F
+from snowflake.snowpark._internal import utils as snowpark_utils
+
+_PROJECT = "ModelDevelopment"
+_SUBPROJECT = "Metrics"
+
+
+@telemetry.send_api_usage_telemetry(project=_PROJECT, subproject=_SUBPROJECT)
+def roc_curve(
+    *,
+    df: snowpark.DataFrame,
+    y_true_col_name: str,
+    y_score_col_name: str,
+    pos_label: Optional[Union[str, int]] = None,
+    sample_weight_col_name: Optional[str] = None,
+    drop_intermediate: bool = True,
+) -> Tuple[npt.ArrayLike, npt.ArrayLike, npt.ArrayLike]:
+    """
+    Compute Receiver operating characteristic (ROC).
+
+    Note: this implementation is restricted to the binary classification task.
+
+    Args:
+        df: Input dataframe.
+        y_true_col_name: Column name representing true binary labels.
+            If labels are not either {-1, 1} or {0, 1}, then pos_label should be
+            explicitly given.
+        y_score_col_name: Column name representing target scores, can either
+            be probability estimates of the positive class, confidence values,
+            or non-thresholded measure of decisions (as returned by
+            "decision_function" on some classifiers).
+        pos_label: The label of the positive class.
+            When ``pos_label=None``, if `y_true` is in {-1, 1} or {0, 1},
+            ``pos_label`` is set to 1, otherwise an error will be raised.
+        sample_weight_col_name: Column name representing sample weights.
+        drop_intermediate: Whether to drop some suboptimal thresholds which would
+            not appear on a plotted ROC curve. This is useful in order to create
+            lighter ROC curves.
+
+    Returns:
+        fpr: ndarray of shape (>2,)
+            Increasing false positive rates such that element i is the false
+            positive rate of predictions with score >= `thresholds[i]`.
+        tpr : ndarray of shape (>2,)
+            Increasing true positive rates such that element `i` is the true
+            positive rate of predictions with score >= `thresholds[i]`.
+        thresholds : ndarray of shape = (n_thresholds,)
+            Decreasing thresholds on the decision function used to compute
+            fpr and tpr. `thresholds[0]` represents no instances being predicted
+            and is arbitrarily set to `max(y_score) + 1`.
+    """
+    session = df._session
+    assert session is not None
+    sproc_name = f"roc_curve_{snowpark_utils.generate_random_alphanumeric()}"
+    statement_params = telemetry.get_statement_params(_PROJECT, _SUBPROJECT)
+
+    cols = [y_true_col_name, y_score_col_name]
+    if sample_weight_col_name:
+        cols.append(sample_weight_col_name)
+    query = df[cols].queries["queries"][-1]
+
+    @F.sproc(  # type: ignore[misc]
+        session=session,
+        name=sproc_name,
+        replace=True,
+        packages=["cloudpickle", "scikit-learn", "snowflake-snowpark-python"],
+        statement_params=statement_params,
+    )
+    def roc_curve_sproc(session: snowpark.Session) -> bytes:
+        df = session.sql(query).to_pandas(statement_params=statement_params)
+        y_true = df[y_true_col_name]
+        y_score = df[y_score_col_name]
+        sample_weight = df[sample_weight_col_name] if sample_weight_col_name else None
+        fpr, tpr, thresholds = metrics.roc_curve(
+            y_true,
+            y_score,
+            pos_label=pos_label,
+            sample_weight=sample_weight,
+            drop_intermediate=drop_intermediate,
+        )
+
+        return cloudpickle.dumps((fpr, tpr, thresholds))  # type: ignore[no-any-return]
+
+    loaded_data = cloudpickle.loads(session.call(sproc_name))
+    res: Tuple[npt.ArrayLike, npt.ArrayLike, npt.ArrayLike] = loaded_data
+    return res

tests/integ/snowflake/ml/modeling/framework/utils.py

@@ -144,7 +144,7 @@ class DataType(Enum):
 
 
 def gen_fuzz_data(
-    rows: int, types: List[DataType], low: int = MIN_INT, high: int = MAX_INT
+    rows: int, types: List[DataType], low: Union[int, List[int]] = MIN_INT, high: Union[int, List[int]] = MAX_INT
 ) -> Tuple[List[Any], List[str]]:
     """
     Generate random data based on input column types and row count.
@@ -153,8 +153,8 @@ def gen_fuzz_data(
     Args:
         rows: num of rows to generate
         types: type per column
-        low: lower bound of the output interval (inclusive)
-        high: upper bound of the output interval (exclusive)
+        low: lower bound(s) of the output interval (inclusive)
+        high: upper bound(s) of the output interval (exclusive)
 
     Returns:
         A tuple of generated data and column names
@@ -166,10 +166,12 @@ def gen_fuzz_data(
     names = ["ID"]
 
     for idx, t in enumerate(types):
+        _low = low if isinstance(low, int) else low[idx]
+        _high = high if isinstance(high, int) else high[idx]
         if t == DataType.INTEGER:
-            data.append(np.random.randint(low, high, rows))
+            data.append(np.random.randint(_low, _high, rows))
         elif t == DataType.FLOAT:
-            data.append(np.random.uniform(low, high, rows))
+            data.append(np.random.uniform(_low, _high, rows))
         else:
             raise ValueError(f"Unsupported data type {t}")
         names.append(f"COL_{idx}")

tests/integ/snowflake/ml/modeling/metrics/BUILD.bazel

@@ -2,6 +2,9 @@ load("//bazel:py_rules.bzl", "py_test")
 
 package(default_visibility = ["//visibility:public"])
 
+SHARD_COUNT = 3
+TIMEOUT = "long"  # 900s
+
 py_test(
     name = "test_r2_score",
     srcs = ["test_r2_score.py"],
@@ -23,7 +26,7 @@ py_test(
 
 py_test(
     name = "test_confusion_matrix",
-    timeout = "long",
+    timeout = TIMEOUT,
     srcs = ["test_confusion_matrix.py"],
     deps = [
         "//snowflake/ml/modeling/metrics",
@@ -34,7 +37,7 @@ py_test(
 
 py_test(
     name = "test_correlation",
-    timeout = "long",
+    timeout = TIMEOUT,
     srcs = ["test_correlation.py"],
     deps = [
         "//snowflake/ml/modeling/metrics",
@@ -44,7 +47,7 @@ py_test(
 
 py_test(
     name = "test_covariance",
-    timeout = "long",
+    timeout = TIMEOUT,
     srcs = ["test_covariance.py"],
     deps = [
         "//snowflake/ml/modeling/metrics",
@@ -54,8 +57,9 @@ py_test(
 
 py_test(
     name = "test_precision_recall_fscore_support",
-    timeout = "long",
+    timeout = TIMEOUT,
     srcs = ["test_precision_recall_fscore_support.py"],
+    shard_count = SHARD_COUNT,
     deps = [
         "//snowflake/ml/modeling/metrics",
         "//snowflake/ml/utils:connection_params",
@@ -65,8 +69,21 @@ py_test(
 
 py_test(
     name = "test_precision_score",
-    timeout = "long",
+    timeout = TIMEOUT,
     srcs = ["test_precision_score.py"],
+    shard_count = SHARD_COUNT,
+    deps = [
+        "//snowflake/ml/modeling/metrics",
+        "//snowflake/ml/utils:connection_params",
+        "//tests/integ/snowflake/ml/modeling/framework:utils",
+    ],
+)
+
+py_test(
+    name = "test_roc_curve",
+    timeout = TIMEOUT,
+    srcs = ["test_roc_curve.py"],
+    shard_count = SHARD_COUNT,
     deps = [
         "//snowflake/ml/modeling/metrics",
         "//snowflake/ml/utils:connection_params",