Raise error if torch-scatter is not installed or wrong version is installed (#2486)

* automatically download correct torch-scatter version

* raise error if torch-scatter is not installed

* Update Documentation & Code Style

* catch all import errors and fix linter

* Update Documentation & Code Style

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

docs/_src/tutorials/tutorials/15.md

@@ -36,7 +36,10 @@ Make sure you enable the GPU runtime to experience decent speed in this tutorial
 !pip install git+https://github.com/deepset-ai/haystack.git#egg=farm-haystack[colab]
 
 # The TaPAs-based TableReader requires the torch-scatter library
-!pip install torch-scatter -f https://data.pyg.org/whl/torch-1.10.0+cu113.html
+import torch
+
+version = torch.__version__
+!pip install torch-scatter -f https://data.pyg.org/whl/torch-{version}.html
 
 # Install pygraphviz for visualization of Pipelines
 !apt install libgraphviz-dev

haystack/nodes/reader/table.py

@@ -21,6 +21,15 @@ from haystack.schema import Document, Answer, Span
 from haystack.nodes.reader.base import BaseReader
 from haystack.modeling.utils import initialize_device_settings
 
+torch_scatter_installed = True
+torch_scatter_wrong_version = False
+try:
+    import torch_scatter  # pylint: disable=unused-import
+except ImportError:
+    torch_scatter_installed = False
+except OSError:
+    torch_scatter_wrong_version = True
+
 
 logger = logging.getLogger(__name__)
 
@@ -95,6 +104,15 @@ class TableReader(BaseReader):
                             query + table exceed max_seq_len, the table will be truncated by removing rows until the
                             input size fits the model.
         """
+        if not torch_scatter_installed:
+            raise ImportError(
+                "Please install torch_scatter to use TableReader. You can follow the instructions here: https://github.com/rusty1s/pytorch_scatter"
+            )
+        if torch_scatter_wrong_version:
+            raise ImportError(
+                "torch_scatter could not be loaded. This could be caused by a mismatch between your cuda version and the one used by torch_scatter."
+                "Please try to reinstall torch-scatter. You can follow the instructions here: https://github.com/rusty1s/pytorch_scatter"
+            )
         super().__init__()
 
         self.devices, _ = initialize_device_settings(use_cuda=use_gpu, multi_gpu=False)

tutorials/Tutorial15_TableQA.ipynb

@@ -55,7 +55,10 @@
     "!pip install git+https://github.com/deepset-ai/haystack.git#egg=farm-haystack[colab]\n",
     "\n",
     "# The TaPAs-based TableReader requires the torch-scatter library\n",
-    "!pip install torch-scatter -f https://data.pyg.org/whl/torch-1.10.0+cu113.html\n",
+    "import torch\n",
+    "\n",
+    "version = torch.__version__\n",
+    "!pip install torch-scatter -f https://data.pyg.org/whl/torch-{version}.html\n",
     "\n",
     "# Install pygraphviz for visualization of Pipelines\n",
     "!apt install libgraphviz-dev\n",