first sbs

Author: xadupre

_unittests/ut_helpers/test_log_helper.py

@@ -470,6 +470,55 @@ def test_historical_cube_time_mask(self):
         cube = CubeLogs(df, keys=["^m_*", "exporter"], time="date").load()
         cube.to_excel(output, views=["time_p"], time_mask=True, verbose=1)
 
+    def test_cube_sbs(self):
+        df = pandas.DataFrame(
+            [
+                dict(
+                    date="2025/01/01",
+                    time_p=0.51,
+                    exporter="E1",
+                    opt="O",
+                    perf=3.7,
+                    m_name="A",
+                    m_cls="CA",
+                ),
+                dict(
+                    date="2025/01/01",
+                    time_p=0.51,
+                    perf=3.4,
+                    exporter="E2",
+                    opt="O",
+                    m_name="A",
+                    m_cls="CA",
+                ),
+                dict(
+                    date="2025/01/01",
+                    time_p=0.71,
+                    perf=3.5,
+                    exporter="E2",
+                    opt="O",
+                    m_name="B",
+                    m_cls="CA",
+                ),
+                dict(
+                    date="2025/01/01",
+                    time_p=0.71,
+                    perf=3.6,
+                    exporter="E2",
+                    opt="K",
+                    m_name="B",
+                    m_cls="CA",
+                ),
+            ]
+        )
+        cube = CubeLogs(
+            df, keys=["^m_*", "exporter", "opt"], values=["time_p", "perf"], time="date"
+        ).load()
+        sbs = cube.sbs([dict(exporter="E1", opt="O"), dict(exporter="E2", opt="O")])
+        self.assertEqual(sbs.shape, (4, 2))
+        self.assertEqual(sbs.index.names, ["METRICS", "m_name"])
+        self.assertEqual(sbs.columns.names, ["exporter"])
+
 
 if __name__ == "__main__":
     unittest.main(verbosity=2)

onnx_diagnostic/helpers/log_helper.py

@@ -1248,6 +1248,43 @@ def cube_time(self, fill_other_dates: bool = False, threshold: float = 1.2) -> "
             )
         return self.clone(data=dgr.reset_index(drop=False))
 
+    def sbs(self, configs: Sequence[Dict[str, Any]]) -> pandas.DataFrame:
+        """
+        Creates a side-by-side for two configurations.
+        Every configuration a dictionary column:value which filters in
+        the rows to keep in order to compute the side by side.
+        """
+        set_keys_time = set(self.keys_time)
+        columns_index = None
+        datas = []
+        for conf in configs:
+            if columns_index is None:
+                columns_index = list(conf.keys())
+                assert (
+                    set(columns_index) <= set_keys_time
+                ), f"Configuration {conf} includes columns outside the keys."
+            else:
+                assert set(columns_index) == set(conf), (
+                    f"Every conf should share the same keys but conf={conf} "
+                    f"is different from {set(columns_index)}"
+                )
+            data = self.data
+            for k, v in conf.items():
+                data = data[data[k] == v]
+            assert data.shape[0] > 0, f"No rows found for conf={conf}"
+            datas.append((conf, data))
+
+        new_data = pandas.concat([d[1] for d in datas], axis=0)
+        cube = self.clone(new_data)
+        key_index = [c for c in self.keys_time if c not in set(columns_index)]
+        view = CubeViewDef(key_index=key_index, name="sbs", values=cube.values)
+        res = cube.view(view)
+        res = res.stack("METRICS", future_stack=True)
+        res = res.reorder_levels(
+            [res.index.nlevels - 1, *list(range(res.index.nlevels - 1))]
+        ).sort_index()
+        return res
+
 
 class CubeLogsPerformance(CubeLogs):
     """