[documentation] adding desc for adaptive_autorange (pytorch#111612)

Summary: This prevented it from showing up in docs

Test Plan: no functional changes

Reviewers:

Subscribers:

Tasks:

Tags:
Pull Request resolved: pytorch#111612
Approved by: https://github.com/cpuhrsch

torch/utils/benchmark/utils/timer.py

@@ -328,36 +328,6 @@ def _estimate_block_size(self, min_run_time: float) -> int:
                 number *= 10
         return number
 
-    def adaptive_autorange(
-            self,
-            threshold: float = 0.1,
-            *,
-            min_run_time: float = 0.01,
-            max_run_time: float = 10.0,
-            callback: Optional[Callable[[int, float], NoReturn]] = None,
-    ) -> common.Measurement:
-        number = self._estimate_block_size(min_run_time=0.05)
-
-        def time_hook() -> float:
-            return self._timeit(number)
-
-        def stop_hook(times: List[float]) -> bool:
-            if len(times) > 3:
-                return common.Measurement(
-                    number_per_run=number,
-                    raw_times=times,
-                    task_spec=self._task_spec
-                ).meets_confidence(threshold=threshold)
-            return False
-        times = self._threaded_measurement_loop(
-            number, time_hook, stop_hook, min_run_time, max_run_time, callback=callback)
-
-        return common.Measurement(
-            number_per_run=number,
-            raw_times=times,
-            task_spec=self._task_spec
-        )
-
     def blocked_autorange(
         self,
         callback: Optional[Callable[[int, float], NoReturn]] = None,
@@ -418,6 +388,69 @@ def stop_hook(times: List[float]) -> bool:
             task_spec=self._task_spec
         )
 
+    def adaptive_autorange(
+            self,
+            threshold: float = 0.1,
+            *,
+            min_run_time: float = 0.01,
+            max_run_time: float = 10.0,
+            callback: Optional[Callable[[int, float], NoReturn]] = None,
+    ) -> common.Measurement:
+        """Similar to `blocked_autorange` but also checks for variablility in measurements
+        and repeats until iqr/median is smaller than `threshold` or `max_run_time` is reached.
+
+
+        At a high level, adaptive_autorange executes the following pseudo-code::
+
+            `setup`
+
+            times = []
+            while times.sum < max_run_time
+                start = timer()
+                for _ in range(block_size):
+                    `stmt`
+                times.append(timer() - start)
+
+                enough_data = len(times)>3 and times.sum > min_run_time
+                small_iqr=times.iqr/times.mean<threshold
+
+                if enough_data and small_iqr:
+                    break
+
+        Args:
+            threshold: value of iqr/median threshold for stopping
+
+            min_run_time: total runtime needed before checking `threshold`
+
+            max_run_time: total runtime  for all measurements regardless of `threshold`
+
+        Returns:
+            A `Measurement` object that contains measured runtimes and
+            repetition counts, and can be used to compute statistics.
+            (mean, median, etc.)
+        """
+        number = self._estimate_block_size(min_run_time=0.05)
+
+        def time_hook() -> float:
+            return self._timeit(number)
+
+        def stop_hook(times: List[float]) -> bool:
+            if len(times) > 3:
+                return common.Measurement(
+                    number_per_run=number,
+                    raw_times=times,
+                    task_spec=self._task_spec
+                ).meets_confidence(threshold=threshold)
+            return False
+        times = self._threaded_measurement_loop(
+            number, time_hook, stop_hook, min_run_time, max_run_time, callback=callback)
+
+        return common.Measurement(
+            number_per_run=number,
+            raw_times=times,
+            task_spec=self._task_spec
+        )
+
     @overload
     def collect_callgrind(
         self,