Merge pull request #254 from google/variable-length-windows

Variable window length

.github/workflows/test.yaml

@@ -39,7 +39,7 @@ jobs:
         run: poetry install --no-interaction
 
       - name: Run tests
-        run: poetry run bazel test //temporian/...:all //docs/...:all //tools/...:all --test_output=errors
+        run: poetry run bazel test //temporian/...:all //docs/...:all //tools/...:all --test_output=errors --compilation_mode=dbg
         # Note: Since mid-August 23, using "//...:all" incorrectly try to
         # compile the directory "build" created during the installation of one
         # of the dependency during "poetry install".

.vscode/settings.json

@@ -30,6 +30,8 @@
   "prettier.tabWidth": 2,
   "editor.detectIndentation": true,
   "files.associations": {
+    "*.excalidraw": "json",
+    "*.excalidrawlib": "json",
     "system_error": "cpp",
     "type_traits": "cpp",
     "xtr1common": "cpp",
@@ -92,7 +94,8 @@
     "random": "cpp",
     "string_view": "cpp",
     "numbers": "cpp",
-    "any": "cpp"
+    "any": "cpp",
+    "bitset": "cpp"
   },
   "python.analysis.typeCheckingMode": "basic",
   "jupyter.notebookFileRoot": "${fileDirname}",

CONTRIBUTING.md

@@ -76,6 +76,8 @@ bazel test //...:all
 
 You can use the Bazel test flag `--test_output=streamed` to see the test logs in realtime.
 
+If developing and testing C++ code, the `--compilation_mode=dbg` flag enables additional assertions that are otherwise disabled.
+
 Note that these tests also include docstring examples, using the builtin `doctest` module.
 See the [Adding code examples](#adding-code-examples) section for more information.
 

docs/src/reference/temporian/internals/typing/WindowLength.md

@@ -0,0 +1 @@
+::: temporian.core.typing.WindowLength

temporian/core/event_set_ops.py

@@ -14,15 +14,19 @@
 
 # pylint: disable=import-outside-toplevel
 
-
 from __future__ import annotations
 from typing import Any, Dict, List, Optional, Union, TYPE_CHECKING
 
 from temporian.core.data.duration import Duration
 
 
 if TYPE_CHECKING:
-    from temporian.core.typing import EventSetOrNode, TypeOrDType, IndexKeyList
+    from temporian.core.typing import (
+        EventSetOrNode,
+        TypeOrDType,
+        IndexKeyList,
+        WindowLength,
+    )
 
 T_SCALAR = (int, float)
 
@@ -1379,17 +1383,20 @@ def leak(self: EventSetOrNode, duration: Duration) -> EventSetOrNode:
 
     def moving_count(
         self: EventSetOrNode,
-        window_length: Duration,
+        window_length: WindowLength,
         sampling: Optional[EventSetOrNode] = None,
     ) -> EventSetOrNode:
         """Gets the number of events in a sliding window.
 
         Create a tp.int32 feature containing the number of events in the time
         window (t - window_length, t].
 
-        If the `sampling` argument is not provided, outputs a timestamp for
-        each timestamp in `input`. If the `sampling` argument is provided,
-        outputs a timestamp for each timestamp in `sampling`.
+        `sampling` can't be  specified if a variable `window_length` is
+        specified (i.e. if `window_length` is an EventSet).
+
+        If `sampling` is specified or `window_length` is an EventSet, the moving
+        window is sampled at each timestamp in them, else it is sampled on the
+        input's.
 
         Example without sampling:
             ```python
@@ -1418,6 +1425,27 @@ def moving_count(
 
             ```
 
+        Example with variable window length:
+            ```python
+            >>> a = tp.event_set(timestamps=[0, 1, 2, 5])
+            >>> b = tp.event_set(
+            ...     timestamps=[0, 3, 3, 3, 9],
+            ...     features={
+            ...         "w": [1, 0.5, 3.5, 2.5, 5],
+            ...     },
+            ... )
+            >>> c = a.moving_count(window_length=b)
+            >>> c
+            indexes: []
+            features: [('count', int32)]
+            events:
+                (5 events):
+                    timestamps: [0. 3. 3. 3. 9.]
+                    'count': [1 0 3 2 1]
+            ...
+
+            ```
+
         Example with index:
             ```python
             >>> a = tp.event_set(
@@ -1459,7 +1487,7 @@ def moving_count(
 
     def moving_max(
         self: EventSetOrNode,
-        window_length: Duration,
+        window_length: WindowLength,
         sampling: Optional[EventSetOrNode] = None,
     ) -> EventSetOrNode:
         """Computes the maximum in a sliding window over an
@@ -1469,11 +1497,16 @@ def moving_max(
         returns at time t the max of non-nan values for the feature in the window
         (t - window_length, t].
 
-        If `sampling` is provided samples the moving window's value at each
-        timestamp in `sampling`, else samples it at each timestamp in the input.
+        `sampling` can't be  specified if a variable `window_length` is
+        specified (i.e. if `window_length` is an EventSet).
 
-        If the window does not contain any values (e.g., all the values are missing,
-        or the window does not contain any sampling), outputs missing values.
+        If `sampling` is specified or `window_length` is an EventSet, the moving
+        window is sampled at each timestamp in them, else it is sampled on the
+        input's.
+
+        If the window does not contain any values (e.g., all the values are
+        missing, or the window does not contain any sampling), outputs missing
+        values.
 
         Example:
             ```python
@@ -1509,7 +1542,7 @@ def moving_max(
 
     def moving_min(
         self: EventSetOrNode,
-        window_length: Duration,
+        window_length: WindowLength,
         sampling: Optional[EventSetOrNode] = None,
     ) -> EventSetOrNode:
         """Computes the minimum of values in a sliding window over an
@@ -1519,11 +1552,16 @@ def moving_min(
         returns at time t the minimum of non-nan values for the feature in the window
         (t - window_length, t].
 
-        If `sampling` is provided samples the moving window's value at each
-        timestamp in `sampling`, else samples it at each timestamp in the input.
+        `sampling` can't be  specified if a variable `window_length` is
+        specified (i.e. if `window_length` is an EventSet).
+
+        If `sampling` is specified or `window_length` is an EventSet, the moving
+        window is sampled at each timestamp in them, else it is sampled on the
+        input's.
 
-        If the window does not contain any values (e.g., all the values are missing,
-        or the window does not contain any sampling), outputs missing values.
+        If the window does not contain any values (e.g., all the values are
+        missing, or the window does not contain any sampling), outputs missing
+        values.
 
         Example:
             ```python
@@ -1542,8 +1580,8 @@ def moving_min(
 
             ```
 
-        See [`EventSet.moving_count()`][temporian.EventSet.moving_count] for examples
-        of moving window operations with external sampling and indices.
+        See [`EventSet.moving_count()`][temporian.EventSet.moving_count] for
+        examples of moving window operations with external sampling and indices.
 
         Args:
             window_length: Sliding window's length.
@@ -1559,23 +1597,28 @@ def moving_min(
 
     def moving_standard_deviation(
         self: EventSetOrNode,
-        window_length: Duration,
+        window_length: WindowLength,
         sampling: Optional[EventSetOrNode] = None,
     ) -> EventSetOrNode:
         """Computes the standard deviation of values in a sliding window over an
         [`EventSet`][temporian.EventSet].
 
-        For each t in sampling, and for each feature independently, returns at time
-        t the standard deviation for the feature in the window
+        For each t in sampling, and for each feature independently, returns at
+        time t the standard deviation for the feature in the window
         (t - window_length, t].
 
-        If `sampling` is provided samples the moving window's value at each
-        timestamp in `sampling`, else samples it at each timestamp in the input.
+        `sampling` can't be  specified if a variable `window_length` is
+        specified (i.e. if `window_length` is an EventSet).
+
+        If `sampling` is specified or `window_length` is an EventSet, the moving
+        window is sampled at each timestamp in them, else it is sampled on the
+        input's.
 
         Missing values (such as NaNs) are ignored.
 
-        If the window does not contain any values (e.g., all the values are missing,
-        or the window does not contain any sampling), outputs missing values.
+        If the window does not contain any values (e.g., all the values are
+        missing, or the window does not contain any sampling), outputs missing
+        values.
 
         Example:
             ```python
@@ -1594,8 +1637,8 @@ def moving_standard_deviation(
 
             ```
 
-        See [`EventSet.moving_count()`][temporian.EventSet.moving_count] for examples of moving window
-        operations with external sampling and indices.
+        See [`EventSet.moving_count()`][temporian.EventSet.moving_count] for
+        examples of moving window operations with external sampling and indices.
 
         Args:
             window_length: Sliding window's length.
@@ -1616,22 +1659,27 @@ def moving_standard_deviation(
 
     def moving_sum(
         self: EventSetOrNode,
-        window_length: Duration,
+        window_length: WindowLength,
         sampling: Optional[EventSetOrNode] = None,
     ) -> EventSetOrNode:
         """Computes the sum of values in a sliding window over an
         [`EventSet`][temporian.EventSet].
 
-        For each t in sampling, and for each feature independently, returns at time
-        t the sum of the feature in the window (t - window_length, t].
+        For each t in sampling, and for each feature independently, returns at
+        time t the sum of the feature in the window (
 
-        If `sampling` is provided samples the moving window's value at each
-        timestamp in `sampling`, else samples it at each timestamp in the input.
+            in them. `sampling` can't be specified if `window_length` is  an EventSet.ta variable  - window_length
+            specified, (i.e. if `window_length` is an EventSet).
+
+        If `sampling` is specified or `window_length` is an EventSet, the moving
+        window is sampled at each timesta, else it is sampled on the
+        input's.p
 
         Missing values (such as NaNs) are ignored.
 
-        If the window does not contain any values (e.g., all the values are missing,
-        or the window does not contain any sampling), outputs missing values.
+        If the window does not contain any values (e.g., all the values are
+        missing, or the window does not contain any sampling), outputs missing
+        values.
 
         Example:
             ```python
@@ -1650,8 +1698,8 @@ def moving_sum(
 
             ```
 
-        See [`EventSet.moving_count()`][temporian.EventSet.moving_count] for examples of moving window
-        operations with external sampling and indices.
+        See [`EventSet.moving_count()`][temporian.EventSet.moving_count] for
+        examples of moving window operations with external sampling and indices.
 
         Args:
             window_length: Sliding window's length.
@@ -2149,37 +2197,44 @@ def set_index(
 
         Args:
             indexes: List of index / feature names (strings) used as
-                the new indexes. These names should be either indexes or features in
-                the input.
+                the new indexes. These names should be either indexes or
+                features in the input.
 
         Returns:
             EventSet with the updated indexes.
 
         Raises:
-            KeyError: If any of the specified `indexes` are not found in the input.
+            KeyError: If any of the specified `indexes` are not found in the
+                input.
         """
         from temporian.core.operators.add_index import set_index
 
         return set_index(self, indexes=indexes)
 
     def simple_moving_average(
         self: EventSetOrNode,
-        window_length: Duration,
+        window_length: WindowLength,
         sampling: Optional[EventSetOrNode] = None,
     ) -> EventSetOrNode:
         """Computes the average of values in a sliding window over an
         [`EventSet`][temporian.EventSet].
 
-        For each t in sampling, and for each feature independently, returns at time
-        t the average value of the feature in the window (t - window_length, t].
+        For each t in sampling, and for each feature independently, returns at
+        time t the average value of the feature in the window
+        (t - window_length, t].
+
+        `sampling` can't be  specified if a variable `window_length` is
+        specified (i.e. if `window_length` is an EventSet).
 
-        If `sampling` is provided samples the moving window's value at each
-        timestamp in `sampling`, else samples it at each timestamp in the input.
+        If `sampling` is specified or `window_length` is an EventSet, the moving
+        window is sampled at each timestamp in them, else it is sampled on the
+        input's.
 
         Missing values (such as NaNs) are ignored.
 
-        If the window does not contain any values (e.g., all the values are missing,
-        or the window does not contain any sampling), outputs missing values.
+        If the window does not contain any values (e.g., all the values are
+        missing, or the window does not contain any timestamp), outputs missing
+        values.
 
         Example:
             ```python
@@ -2198,8 +2253,8 @@ def simple_moving_average(
 
             ```
 
-        See [`EventSet.moving_count()`][temporian.EventSet.moving_count] for examples of moving window
-        operations with external sampling and indices.
+        See [`EventSet.moving_count()`][temporian.EventSet.moving_count] for
+        examples of moving window operations with external sampling and indices.
 
         Args:
             window_length: Sliding window's length.

temporian/core/operators/window/BUILD

@@ -25,6 +25,7 @@ py_library(
     srcs = ["base.py"],
     srcs_version = "PY3",
     deps = [
+        "//temporian/core:typing",
         "//temporian/core/data:dtype",
         "//temporian/core/data:duration_utils",
         "//temporian/core/data:node",