Add inactive sensors to graph

Created by: RasmusOrsoe

This PR is a draft. It's a bit rough around the edges but outlines a method of adding inactive sensor modules to the graphs.

The general idea is to add abstract (backend specific implementations needed) methods to the Dataset class that queries geometry_tables for sensor_id's. In SQLite I've created a method that constructs geometry_table (look-up table for geometry) from pulsemap tables.

Right now, the _get_inactive_sensors will actually extract the active pulsemap on its own (and create the UID in the process). This means that every pulsemap will be extracted twice; a bit inefficient.

Let me know what you think of the general approach! I'll try to polish this up during the week.

src/graphnet/data/sqlite/sqlite_dataset.py

@@ -3,9 +3,12 @@
 from typing import Any, List, Optional, Tuple, Union
 import pandas as pd
 import sqlite3
+import numpy as np
 
 from graphnet.data.dataset import Dataset, ColumnMissingException
 
+from graphnet.data.sqlite.sqlite_utilities import database_table_exists
+
 
 class SQLiteDataset(Dataset):
     """Pytorch dataset for reading data from SQLite databases."""
@@ -67,11 +70,10 @@ class SQLiteDataset(Dataset):
                 combined_selections = (
                     f"{self._index_column} = {index} and {selection}"
                 )
-
-            result = self._conn.execute(
-                f"SELECT {columns} FROM {table} WHERE "
-                f"{combined_selections}"
-            ).fetchall()
+                result = self._conn.execute(
+                    f"SELECT {columns} FROM {table} WHERE "
+                    f"{combined_selections}"
+                ).fetchall()
         except sqlite3.OperationalError as e:
             if "no such column" in str(e):
                 raise ColumnMissingException(str(e))
@@ -79,6 +81,23 @@ class SQLiteDataset(Dataset):
                 raise e
         return result
 
+    def _get_inactive_sensors(
+        self,
+        features: List[Tuple[float, ...]],
+        columns: List[str],
+        sequential_index: int,
+    ) -> List[Tuple[float, ...]]:
+        assert self._conn
+        index = self._get_event_index(sequential_index)
+        active_query = f"select (CAST({self._sensor_position['x']} AS str) || '_' || CAST({self._sensor_position['y']} AS str) || '_' || CAST({self._sensor_position['z']} AS str)) as UID from {self._pulsemaps[0]} where {self._index_column} = {index} and {self._selection}"
+        active_result = self._conn.execute(active_query).fetchall()
+        columns_str = ", ".join(
+            columns
+        )  # remove event_no because not in geometry table
+        query = f"select {columns_str} from {self._geometry_table} where UID not in {str(tuple(np.array(active_result)))}"
+        inactive_result = self._conn.execute(query).fetchall()
+        return inactive_result
+
     def _get_all_indices(self) -> List[int]:
         self._establish_connection(0)
         indices = pd.read_sql_query(
@@ -147,3 +166,15 @@ class SQLiteDataset(Dataset):
                 self._all_connections_established = False
                 self._conn = None
         return self
+
+    def _setup_geometry_table(self, geometry_table: str) -> None:
+        """Assign the geometry table to self if it exists in the database."""
+        assert isinstance(self._path, str)
+        if database_table_exists(
+            database_path=self._path, table_name=geometry_table
+        ):
+            self._geoemtry_table = geometry_table
+        else:
+            assert (
+                1 == 2
+            ), f"Geometry table named {geometry_table} is not in the database {self._path}"