Merge branch 'develop' into r0.5

intel · Jun 23, 2023 · ba89871 · ba89871
2 parents 9b74b91 + d767ac8
commit ba89871
Show file tree

Hide file tree

Showing 11 changed files with 77 additions and 83 deletions.
diff --git a/explainer/attributions/attributions.py b/explainer/attributions/attributions.py
@@ -215,11 +215,11 @@ def __init__(self,
         self.shap_values = self.explainer.shap_values(self.targets, nsamples=nsamples)
         self.info_panel = force_plot_info_panel
 
-    def visualize(self) -> None:
+    def visualize(self):
         '''
         Display the force plot of the of the target example(s)
         '''
-        self.force_plot(self.explainer.expected_value, self.shap_values[0], self.targets)
+        return self.force_plot(self.explainer.expected_value, self.shap_values[0], self.targets)
 
 
 class PartitionExplainer(FeatureAttributions):

diff --git a/notebooks/explainer/cifar_with_attributions/TorchVision_CIFAR_Interpret.ipynb b/notebooks/explainer/cifar_with_attributions/TorchVision_CIFAR_Interpret.ipynb
@@ -7,6 +7,19 @@
     "# Explaining Custom CNN CIFAR-10 Classification Using the Attributions Explainer"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# temp workaround for captum bug\n",
+    "# issue on captum: https://github.com/pytorch/captum/issues/1114\n",
+    "# Note: this matplotlib version is not guaranteed to work with any other application of intel-xai\n",
+    "\n",
+    "! pip install -U matplotlib==3.6.3 --no-deps"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -206,6 +219,15 @@
     "net.eval()"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import matplotlib"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -231,13 +253,6 @@
     "attributions.smoothgrad(net).visualize(input,labels[ind],original_image,\"Smooth Grad\")\n",
     "attributions.featureablation(net).visualize(input,labels[ind],original_image,\"Feature Ablation\")"
    ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": []
   }
  ],
  "metadata": {
@@ -256,7 +271,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.9.6"
+   "version": "3.9.16"
   },
   "vscode": {
    "interpreter": {

diff --git a/notebooks/explainer/heart_disease_with_attributions/heart_disease.ipynb b/notebooks/explainer/heart_disease_with_attributions/heart_disease.ipynb
@@ -240,13 +240,6 @@
     "ke = attributions.kernel_explainer(model, X_train.iloc[1:101, :], X_train.iloc[0, :])\n",
     "ke.visualize()"
    ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": []
   }
  ],
  "metadata": {
@@ -265,7 +258,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.9.6"
+   "version": "3.9.16"
   },
   "vscode": {
    "interpreter": {

diff --git a/notebooks/explainer/mnist_with_attributions_and_metrics/mnist.ipynb b/notebooks/explainer/mnist_with_attributions_and_metrics/mnist.ipynb
@@ -272,7 +272,7 @@
    "source": [
     "# instatiate gradient explainer object\n",
     "# run the deep explainer\n",
-    "grViz = attributions.gradient_explainer(model, X_test[:100],  X_test[matches[:6]], 2, classes)\n",
+    "grViz = attributions.gradient_explainer(model, X_test[:100],  X_test[matches[:6]], classes, 2)\n",
     "grViz.visualize()"
    ]
   },
@@ -309,7 +309,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.9.6"
+   "version": "3.9.16"
   }
  },
  "nbformat": 4,

diff --git a/notebooks/explainer/multimodal_cancer_detection/Multimodal_Cancer_Detection.ipynb b/notebooks/explainer/multimodal_cancer_detection/Multimodal_Cancer_Detection.ipynb
@@ -16,7 +16,9 @@
    "cell_type": "code",
    "execution_count": null,
    "id": "ad8a9723-5dbe-44eb-9baa-eca188e435f2",
-   "metadata": {},
+   "metadata": {
+    "scrolled": true
+   },
    "outputs": [],
    "source": [
     "import numpy as np\n",
@@ -705,29 +707,29 @@
     "from explainer import cam\n",
     "final_image_dim = (224, 224)\n",
     "targetLayer = viz_model._model.layer4\n",
-    "xgradcam = cam.xgradcam(viz_model._model, targetLayer, \n",
+    "xgc = cam.x_gradcam(viz_model._model, targetLayer, \n",
     "                      label_map_func('Normal'), \n",
     "                      images[0],\n",
     "                      final_image_dim,\n",
     "                      'cpu')\n",
     "\n",
-    "xgradcam.visualize()\n",
+    "xgc.visualize()\n",
     "\n",
-    "xgradcam = cam.xgradcam(viz_model._model, targetLayer, \n",
+    "xgc = cam.x_gradcam(viz_model._model, targetLayer, \n",
     "                      label_map_func('Normal'), \n",
     "                      images[1],\n",
     "                      final_image_dim,\n",
     "                      'cpu')\n",
     "\n",
-    "xgradcam.visualize()\n",
+    "xgc.visualize()\n",
     "\n",
-    "xgradcam = cam.xgradcam(viz_model._model, targetLayer, \n",
+    "xgc = cam.x_gradcam(viz_model._model, targetLayer, \n",
     "                      label_map_func('Normal'), \n",
     "                      images[2],\n",
     "                      final_image_dim,\n",
     "                      'cpu')\n",
     "\n",
-    "xgradcam.visualize()"
+    "xgc.visualize()"
    ]
   },
   {
@@ -777,30 +779,29 @@
     "\n",
     "final_image_dim = (224, 224)\n",
     "targetLayer = viz_model._model.layer4\n",
-    "#targetLayer = \"_IPEXConv2d-169\"\n",
-    "xgradcam = cam.xgradcam(viz_model._model, targetLayer, \n",
+    "xgc = cam.x_gradcam(viz_model._model, targetLayer, \n",
     "                      label_map_func('Benign'), \n",
     "                      images[0],\n",
     "                      final_image_dim,\n",
     "                      'cpu')\n",
     "\n",
-    "xgradcam.visualize()\n",
+    "xgc.visualize()\n",
     "\n",
-    "xgradcam = cam.xgradcam(viz_model._model, targetLayer, \n",
+    "xgc = cam.x_gradcam(viz_model._model, targetLayer, \n",
     "                      label_map_func('Benign'), \n",
     "                      images[1],\n",
     "                      final_image_dim,\n",
     "                      'cpu')\n",
     "\n",
-    "xgradcam.visualize()\n",
+    "xgc.visualize()\n",
     "\n",
-    "xgradcam = cam.xgradcam(viz_model._model, targetLayer, \n",
+    "xgc = cam.x_gradcam(viz_model._model, targetLayer, \n",
     "                      label_map_func('Benign'), \n",
     "                      images[2],\n",
     "                      final_image_dim,\n",
     "                      'cpu')\n",
     "\n",
-    "xgradcam.visualize()"
+    "xgc.visualize()"
    ]
   },
   {
@@ -1121,7 +1122,7 @@
    "source": [
     "import transformers\n",
     "transformers.set_seed(1)\n",
-    "nlp_history = nlp_model.train(train_nlp_dataset, output_dir, epochs=3, use_trainer=True)"
+    "nlp_history = nlp_model.train(train_nlp_dataset, output_dir, epochs=3, use_trainer=True, seed=1)"
    ]
   },
   {
@@ -1281,7 +1282,7 @@
    "outputs": [],
    "source": [
     "from explainer import attributions\n",
-    "partition_explainer = attributions.partition_explainer(f, r\"\\W+\", test_nlp_dataset.class_names)(np.array(mal_classified_as_ben_text))\n",
+    "partition_explainer = attributions.partition_text_explainer(f, test_nlp_dataset.class_names, np.array(mal_classified_as_ben_text), r\"\\W+\")\n",
     "partition_explainer.visualize()"
    ]
   },
@@ -1305,8 +1306,8 @@
    "outputs": [],
    "source": [
     "from intel_extension_for_transformers.optimization.trainer import NLPTrainer\n",
-    "from intel_extension_for_transformers import objectives, OptimizedModel, QuantizationConfig\n",
-    "from intel_extension_for_transformers import metrics as nlptk_metrics"
+    "from intel_extension_for_transformers.optimization import objectives, OptimizedModel, QuantizationConfig\n",
+    "from intel_extension_for_transformers.optimization import metrics as nlptk_metrics"
    ]
   },
   {
@@ -1389,7 +1390,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "quantized_model.save(os.path.join(output_dir, 'quantized_BERT'))\n",
+    "quantizer.save_model(os.path.join(output_dir, 'quantized_BERT'))\n",
     "nlp_model._model.config.save_pretrained(os.path.join(output_dir, 'quantized_BERT'))"
    ]
   },
@@ -1504,7 +1505,7 @@
     "viz_weight = test_viz_metrics[1]\n",
     "\n",
     "# final weight of nlp is its overall validation accuracy\n",
-    "nlp_weight = eval_acc\n",
+    "nlp_weight = test_nlp_metrics['eval_accuracy']\n",
     "\n",
     "def convert_nomenclature(df_pid):\n",
     "    return 'P' + df_pid[:-1] + '_' + df_pid[-1]\n",
@@ -1597,7 +1598,7 @@
     "oh_y_pred = np.eye(n_values)[y_pred]\n",
     "y_true = [label_map_func(i) for i in label]\n",
     "\n",
-    "ensemble_cm = metrics.confusion_matrix(oh_y_true, oh_y_pred, test_nlp_dataset.class_names)\n",
+    "ensemble_cm = metrics.confusion_matrix(y_true, oh_y_pred, test_nlp_dataset.class_names)\n",
     "ensemble_cm.visualize()\n",
     "print(ensemble_cm.report)"
    ]
@@ -1640,42 +1641,40 @@
     "final_image_dim = (224, 224)\n",
     "targetLayer = viz_model._model.layer4\n",
     "#targetLayer = \"_IPEXConv2d-169\"\n",
-    "xgradcam = cam.xgradcam(viz_model._model, targetLayer, \n",
+    "xgc = cam.x_gradcam(viz_model._model, targetLayer, \n",
     "                      label_map_func('Malignant'), \n",
     "                      images[0],\n",
     "                      final_image_dim,\n",
     "                      'cpu')\n",
     "\n",
-    "xgradcam.visualize()\n",
+    "xgc.visualize()\n",
     "\n",
-    "xgradcam = cam.xgradcam(viz_model._model, targetLayer, \n",
+    "xgc = cam.x_gradcam(viz_model._model, targetLayer, \n",
     "                      label_map_func('Malignant'), \n",
     "                      images[1],\n",
     "                      final_image_dim,\n",
     "                      'cpu')\n",
     "\n",
-    "xgradcam.visualize()\n",
+    "xgc.visualize()\n",
     "\n",
-    "xgradcam = cam.xgradcam(viz_model._model, targetLayer, \n",
+    "xgc = cam.x_gradcam(viz_model._model, targetLayer, \n",
     "                      label_map_func('Malignant'), \n",
     "                      images[2],\n",
     "                      final_image_dim,\n",
     "                      'cpu')\n",
     "\n",
-    "xgradcam.visualize()\n",
+    "xgc.visualize()\n",
     "\n",
-    "xgradcam = cam.xgradcam(viz_model._model, targetLayer, \n",
+    "xgc = cam.x_gradcam(viz_model._model, targetLayer, \n",
     "                      label_map_func('Malignant'), \n",
     "                      images[3],\n",
     "                      final_image_dim,\n",
     "                      'cpu')\n",
     "\n",
-    "xgradcam.visualize()\n",
+    "xgc.visualize()\n",
     "text_for_shap = np.expand_dims(np.array(ensemble_results.iloc[image_idx]['text']), axis=0)\n",
-    "ensemble_partition_explainer = attributions.partition_explainer(f, r\"\\W+\", test_nlp_dataset.class_names)\\\n",
-    "                                                (text_for_shap)\n",
-    "ensemble_partition_explainer.visualize()\n",
-    "\n"
+    "ensemble_partition_explainer = attributions.partition_text_explainer(f,  test_nlp_dataset.class_names, text_for_shap, r\"\\W+\")\n",
+    "ensemble_partition_explainer.visualize()"
    ]
   },
   {

diff --git a/notebooks/explainer/multimodal_cancer_detection/README.md b/notebooks/explainer/multimodal_cancer_detection/README.md
@@ -37,11 +37,9 @@ The `dataset_utils.py` holds the supporting functions that prepare the image and
 
 To run `Multimodal_Cancer_Detection.ipynb`, install the following dependencies:
 1. [Intel® Explainable AI](https://github.com/IntelAI/intel-xai-tools)
-2. [Intel® Transfer Learning Tool](https://github.com/IntelAI/transfer-learning)
+2. `pip install intel-transfer-learning-tool==0.5`
 3. `pip install intel-extension-for-transformers`
 4. `pip install scikit-image`
-5. `pip install jupyterlab`
-6. `pip install jupyter-dash`
 7. `pip install nltk`
 8. `pip install docx2txt`
 9. `pip install openpyxl`

diff --git a/notebooks/explainer/newsgroups_with_attributions_and_metrics/README.md b/notebooks/explainer/newsgroups_with_attributions_and_metrics/README.md
@@ -16,7 +16,6 @@ This notebook demonstrates how to use the attributions explainer API to explain
 
 To run `partitionexplainer.ipynb`, install the following dependencies:
 1. [Intel® Explainable AI](https://github.com/IntelAI/intel-xai-tools)
-2. pip install jupyter-dash
 
 ## References
 

diff --git a/notebooks/explainer/newsgroups_with_attributions_and_metrics/partitionexplainer.ipynb b/notebooks/explainer/newsgroups_with_attributions_and_metrics/partitionexplainer.ipynb
@@ -273,7 +273,7 @@
     "    preds = model.predict(X_batch)\n",
     "    return preds\n",
     "\n",
-    "partition_explainer = attributions.partition_explainer(make_predictions, r\"\\W+\", selected_categories)(X_batch_text)"
+    "partition_explainer = attributions.partition_text_explainer(make_predictions, selected_categories, X_batch_text, r\"\\W+\")"
    ]
   },
   {
@@ -468,7 +468,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.9.6"
+   "version": "3.9.16"
   }
  },
  "nbformat": 4,

diff --git a/..._learning_text_classification/PyTorch_Text_Classifier_fine_tuning_with_Attributions.ipynb b/..._learning_text_classification/PyTorch_Text_Classifier_fine_tuning_with_Attributions.ipynb
@@ -490,6 +490,7 @@
     "\n",
     "        # If training_args are given, we use the `Trainer` API to train the model\n",
     "        if self.training_args:\n",
+    "            self.model.train()\n",
     "            self.trainer = Trainer(model=self.model,\n",
     "                                   args=self.training_args,\n",
     "                                   train_dataset=self.train_ds,\n",
@@ -534,6 +535,7 @@
     "    def evaluate(self, batch_size=16):\n",
     "        \n",
     "        if self.trainer:\n",
+    "            self.model.eval()\n",
     "            metrics = self.trainer.evaluate()\n",
     "            for key in metrics.keys():\n",
     "                print(\"{}: {}\".format(key, metrics[key]))\n",
@@ -872,8 +874,7 @@
     "from explainer import attributions\n",
     "# Get shap values\n",
     "text_for_shap = dataset.dataset['test'][:10]['text']\n",
-    "partition_explainer = attributions.partition_explainer(f, r\"\\W+\", dataset.class_labels.names)\n",
-    "partition_explainer(text_for_shap)"
+    "partition_explainer = attributions.partition_text_explainer(f, dataset.class_labels.names, text_for_shap, r\"\\W+\", )"
    ]
   },
   {

diff --git a/notebooks/explainer/transfer_learning_text_classification/README.md b/notebooks/explainer/transfer_learning_text_classification/README.md
@@ -20,7 +20,10 @@ The notebook performs the following steps:
 
 ## Running the notebook
 
-To run the notebook, follow the instructions to setup the [PyTorch notebook environment](/notebooks#pytorch-environment).
+
+To run `PyTorch_Text_Classifier_fine_tuning_with_Attributions.ipynb`, install the following dependencies:
+1. [Intel® Explainable AI](https://github.com/IntelAI/intel-xai-tools)
+2. `pip install intel-transfer-learning-tool==0.5`
 
 ## References