feat: add sample function (#447)

* add sample function

* `eval_only` supports summarization

python/qianfan/dataset/data_source/chunk_reader.py

@@ -107,7 +107,7 @@ def __init__(
         self.file_path = file_path
         self.fd = open(file_path, mode="r", encoding=encoding())
 
-        self.ijson_object = ijson.items(self.fd, element_json_path)
+        self.ijson_object = ijson.items(self.fd, element_json_path, use_float=True)
 
     def _get_an_element(self, index: int) -> Any:
         return next(self.ijson_object)

python/qianfan/dataset/dataset.py

@@ -1028,6 +1028,36 @@ def take_slice(
         """
         return super().take_slice(start, end, should_create_new_obj, **kwargs)
 
+    @_online_except_decorator
+    def sample(
+        self,
+        sample_number: int,
+        start: int = 0,
+        end: int = -1,
+        should_create_new_obj: bool = False,
+        **kwargs: Any,
+    ) -> Self:
+        """
+        take random slice in dataset
+
+        Args:
+            sample_number (int):
+                how many entries should be sampled
+            start (int):
+                where the sample part starts
+            end (int):
+                where the sample part ends
+            should_create_new_obj (bool):
+                should a new object be created when mapping terminates.
+                Default to False. In some cases, you may want to set
+                this value to True
+            **kwargs (Any):
+                other arguments
+        """
+        return super().sample(
+            sample_number, start, end, should_create_new_obj, **kwargs
+        )
+
     def __getitem__(self, key: Any) -> Any:
         if (
             isinstance(key, int)
@@ -2020,8 +2050,6 @@ def show_processed_statistics(
                 MeanMethod(),
                 MinMethod(),
                 MaxMethod(),
-                QuantileMethod(q=0.2),
-                QuantileMethod(q=0.5),
                 QuantileMethod(q=0.8),
                 QuantileMethod(q=0.9),
             ]

python/qianfan/dataset/table.py

@@ -14,6 +14,7 @@
 """
 wrapper for pyarrow.Table
 """
+import random
 from typing import (
     Any,
     Callable,
@@ -519,6 +520,38 @@ def take_slice(self, start: int = 0, end: int = -1) -> PyarrowTable:
 
         return self.table.slice(start, end - start + 1)
 
+    def sample(
+        self,
+        sample_number: int,
+        start: int = 0,
+        end: int = -1,
+    ) -> PyarrowTable:
+        if start < 0:
+            err_msg = f"start index is smaller than 0: {start}"
+            log_error(err_msg)
+            raise ValueError(err_msg)
+
+        if end >= self.table.num_rows:
+            err_msg = (
+                f"end index {end} is bigger than table size: {self.table.num_rows}"
+            )
+            log_error(err_msg)
+            raise ValueError(err_msg)
+
+        if end < 0:
+            end = self.table.num_rows - 1
+
+        if sample_number < 0:
+            err_msg = f"can't sample {sample_number} entries"
+            log_error(err_msg)
+            raise ValueError(err_msg)
+
+        if sample_number >= self.table.num_rows:
+            return self.table
+
+        numbers = random.sample(range(start, end + 1), sample_number)
+        return self.table.take(numbers)
+
 
 class _PyarrowColumnManipulator(BaseModel, Addable, Listable, Processable):
     """handler for processing of pyarrow table column"""
@@ -1193,6 +1226,18 @@ def take_slice(
         result_ds = manipulator.take_slice(start, end)
         return self._create_new_obj(result_ds, should_create_new_obj)
 
+    def sample(
+        self,
+        sample_number: int,
+        start: int = 0,
+        end: int = -1,
+        should_create_new_obj: bool = False,
+        **kwargs: Any,
+    ) -> Self:
+        manipulator = self._row_op()
+        result_ds = manipulator.sample(sample_number, start, end)
+        return self._create_new_obj(result_ds, should_create_new_obj)
+
     def col_map(
         self,
         op: Callable[[Any], Any],

python/qianfan/evaluation/evaluation_manager.py

@@ -421,7 +421,20 @@ def eval_only(
         tmp_ds = Dataset.create_from_pyobj(
             self._run_evaluator_locally(dataset, **kwargs)
         )
-        return EvaluationResult(result_dataset=dataset.col_append(tmp_ds.col_list()))
+
+        assert self.local_evaluators
+
+        summarization_dict: Dict[str, Any] = {}
+
+        for evaluator in self.local_evaluators:
+            summarization = evaluator.summarize(tmp_ds)
+            if summarization:
+                summarization_dict.update(summarization)
+
+        return EvaluationResult(
+            metrics=summarization_dict,
+            result_dataset=dataset.col_append(tmp_ds.col_list()),
+        )
 
     def eval(
         self,