make skl2onnx optional

The support for scikit-learn pipelines is optional, so we raise
an error only if one tries to use it while skl2onnx is not
installed.

Fixes #21

Author: MainRo

ebm2onnx/sklearn.py

@@ -1,44 +1,54 @@
-from skl2onnx.common.data_types import Int64TensorType, FloatTensorType, StringTensorType
 from . import context
 from . import convert
 
 import onnx
 
 
-def ebm_output_shape_calculator(operator):
-    op = operator.raw_operator
+try:
+    from skl2onnx.common.data_types import Int64TensorType, FloatTensorType, StringTensorType
 
-    operator.outputs[0].type = Int64TensorType([None])  # label
-    operator.outputs[1].type = FloatTensorType([None, len(op.classes_)])  # probabilities
 
+    def ebm_output_shape_calculator(operator):
+        op = operator.raw_operator
 
-def convert_ebm_classifier(scope, operator, container):
-    """Converts an EBM model to ONNX with sklearn-onnx
-    """
-    op = operator.raw_operator
+        operator.outputs[0].type = Int64TensorType([None])  # label
+        operator.outputs[1].type = FloatTensorType([None, len(op.classes_)])  # probabilities
 
-    input_name = operator.inputs[0].onnx_name
-    ctx = context.create(
-        generate_variable_name=scope.get_unique_variable_name,
-        generate_operator_name=scope.get_unique_operator_name,
-    )
 
-    g = convert.to_graph(
-        op, dtype=(input_name, 'float'),
-        name="ebm",
-        predict_proba=True,
-        prediction_name="label",
-        probabilities_name="probabilities",
-        context=ctx
-    )
+    def convert_ebm_classifier(scope, operator, container):
+        """Converts an EBM model to ONNX with sklearn-onnx
+        """
+        op = operator.raw_operator
 
-    for node in g.nodes:
-        v = container._get_op_version(node.domain, node.op_type)
-        container.node_domain_version_pair_sets.add((node.domain, v))
+        input_name = operator.inputs[0].onnx_name
+        ctx = context.create(
+            generate_variable_name=scope.get_unique_variable_name,
+            generate_operator_name=scope.get_unique_operator_name,
+        )
 
-    container.nodes.extend(g.nodes)
+        g = convert.to_graph(
+            op, dtype=(input_name, 'float'),
+            name="ebm",
+            predict_proba=True,
+            prediction_name="label",
+            probabilities_name="probabilities",
+            context=ctx
+        )
 
-    for i in g.initializers:
-        content = i.SerializeToString()
-        container.initializers_strings[content] = i.name
-        container.initializers.append(i)
+        for node in g.nodes:
+            v = container._get_op_version(node.domain, node.op_type)
+            container.node_domain_version_pair_sets.add((node.domain, v))
+
+        container.nodes.extend(g.nodes)
+
+        for i in g.initializers:
+            content = i.SerializeToString()
+            container.initializers_strings[content] = i.name
+            container.initializers.append(i)
+
+except Exception:
+    def ebm_output_shape_calculator(operator):
+        raise ImportError('skl2onnx not found. Please install it to use serialize a model via scikit-learn')
+
+    def convert_ebm_classifier(scope, operator, container):
+        raise ImportError('skl2onnx not found. Please install it to use serialize a model via scikit-learn')

examples/convert.ipynb

@@ -1,5 +1,23 @@
 {
  "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "fd9863cf-daee-4d80-8b1c-2b6fe086b81d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "\"\"\" Uncomment this if your environment does not have all runtime dependencies installed\n",
+    "import pip\n",
+    "pip.main([\n",
+    "    'install',\n",
+    "    'interpret-core[notebook,dash,ploly]',\n",
+    "    'onnxruntime',\n",
+    "    'matplotlib',\n",
+    "])\n",
+    "\"\"\""
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -15,6 +33,19 @@
     "from sklearn.metrics import precision_recall_fscore_support"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "4441327e-f892-4e52-aa1c-c92e43a45696",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from interpret.provider import InlineProvider\n",
+    "from interpret import set_visualize_provider\n",
+    "\n",
+    "set_visualize_provider(InlineProvider())"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -38,6 +69,27 @@
     "# Binary classification"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "01be0d4d-c47a-47a9-b953-b5b7c6a9a67a",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from onnx import defs"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "020a4e17-cbaf-4159-b1e5-11e42b33f49f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#rt.get_available_providers() \n",
+    "onnx.__version__\n"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "powerful-desktop",
@@ -92,9 +144,7 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "canadian-telephone",
-   "metadata": {
-    "scrolled": false
-   },
+   "metadata": {},
    "outputs": [],
    "source": [
     "# A lookup at the generated model\n",
@@ -121,9 +171,30 @@
     "    model=ebm,\n",
     "    dtype=ebm2onnx.get_dtype_from_pandas(x_train),\n",
     "    name=\"ebm\",\n",
+    "    #target_opset=10,\n",
     ")"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "8e40b3d0-c8c7-4178-8c10-826a3d1363ca",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "onnx_model.ir_version = 10"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "bb80a006-8542-485e-a0fa-3473ac214018",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "onnx_model"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "italic-authorization",
@@ -190,7 +261,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.10"
+   "version": "3.13.8"
   }
  },
  "nbformat": 4,