Add explanation notebooks (kubeflow#509)

* Update explainer demos with new notebooks

* Add updated explanation notebooks

* Update from review comments

* clean up

* rerun income explainer notebook to see output

* removing output as plot.ly graphs don't show on github

docs/samples/explanation/alibi/alibi_helper.py

@@ -0,0 +1,76 @@
+import numpy as np
+import requests
+from alibi.datasets import fetch_adult
+import pandas as pd
+import plotly.graph_objects as go
+from IPython.display import display, Markdown, display
+
+def getFeatures(X,cmap):
+    return pd.DataFrame(X).replace(cmap).values.squeeze().tolist()
+
+def predict(X, name, ds, svc_hostname, cluster_ip):
+    formData = {
+    'instances': X
+    }
+    headers = {}
+    headers["Host"] = svc_hostname
+    res = requests.post('http://'+cluster_ip+'/v1/models/'+name+':predict', json=formData, headers=headers)
+    if res.status_code == 200:
+        return ds.target_names[np.array(res.json()["predictions"])[0]]
+    else:
+        print("Failed with ",res.status_code)
+        return []
+
+def explain(X, name, svc_hostname, cluster_ip):
+    formData = {
+    'instances': X
+    }
+    headers = {}
+    headers["Host"] = svc_hostname
+    res = requests.post('http://'+cluster_ip+'/v1/models/'+name+':explain', json=formData, headers=headers)
+    if res.status_code == 200:
+        return res.json()
+    else:
+        print("Failed with ",res.status_code)
+        return []
+
+def show_bar(X, labels, title):
+    fig = go.Figure(go.Bar(x=X,y=labels,orientation='h',width=[0.5]))
+    fig.update_layout(autosize=False,width=700,height=300,
+                      xaxis=dict(range=[0, 1]),
+                      title_text=title,  
+                      font=dict(family="Courier New, monospace",size=18,color="#7f7f7f"
+    ))
+    fig.show()
+
+
+def show_feature_coverage(exp):
+    data = []
+    for idx, name in enumerate(exp["names"]):
+        data.append(go.Bar(name=name, x=["coverage"], y=[exp['raw']['coverage'][idx]]))
+    fig = go.Figure(data=data)
+    fig.update_layout(yaxis=dict(range=[0, 1]))
+    fig.show()
+
+def show_anchors(names):
+    display(Markdown('# Explanation:'))
+    display(Markdown('## {}'.format(names)))
+
+def show_examples(exp,fidx,ds,covered=True):
+    if covered:
+        cname = 'covered'
+        display(Markdown("## Examples covered by Anchors: {}".format(exp['names'][0:fidx+1])))
+    else:
+        cname = 'covered_false'
+        display(Markdown("## Examples not covered by Anchors: {}".format(exp['names'][0:fidx+1])))
+    if "feature_names" in ds:
+        return pd.DataFrame(exp['raw']['examples'][fidx][cname],columns=ds.feature_names)
+    else:
+        return pd.DataFrame(exp['raw']['examples'][fidx][cname])
+
+def show_prediction(prediction):
+    display(Markdown('## Prediction: {}'.format(prediction)))
+
+def show_row(X,ds):
+    display(pd.DataFrame(X,columns=ds.feature_names))
+

docs/samples/explanation/alibi/income/README.md

@@ -2,6 +2,8 @@
 
 This example uses a [US income dataset](https://archive.ics.uci.edu/ml/datasets/adult)
 
+You can also try out the [Jupyter notebook](income_explanations.ipynb).
+
 We can create a InferenceService with a trained sklearn predictor for this dataset and an associated model explainer. The black box explainer algorithm we will use is the Tabular version of Anchors from the [Alibi open source library](https://github.com/SeldonIO/alibi). More details on this algorithm and configuration settings that can be set can be found in the [Seldon Alibi documentation](https://docs.seldon.io/projects/alibi/en/stable/).
 
 The InferenceService is shown below:
@@ -14,19 +16,24 @@ metadata:
 spec:
   default:
     predictor:
+      minReplicas: 1
       sklearn:
         storageUri: "gs://seldon-models/sklearn/income/model"
         resources:
           requests:
             cpu: 0.1
+          limits:
+            cpu: 1
     explainer:
+      minReplicas: 1
       alibi:
         type: AnchorTabular
         storageUri: "gs://seldon-models/sklearn/income/explainer"
         resources:
           requests:
             cpu: 0.1
-        
+          limits:
+            cpu: 1
 ```
 
 Create this InferenceService:

docs/samples/explanation/alibi/income/income.yaml

@@ -2,20 +2,28 @@ apiVersion: "serving.kubeflow.org/v1alpha2"
 kind: "InferenceService"
 metadata:
   name: "income"
-  namespace: default
 spec:
   default:
     predictor:
+      minReplicas: 1
       sklearn:
         storageUri: "gs://seldon-models/sklearn/income/model"
         resources:
           requests:
             cpu: 0.1
+            memory: 1Gi
+          limits:
+            cpu: 1
+            memory: 1Gi            
     explainer:
+      minReplicas: 1
       alibi:
         type: AnchorTabular
         storageUri: "gs://seldon-models/sklearn/income/explainer"
         resources:
           requests:
             cpu: 0.1
-
+            memory: 1Gi
+          limits:
+            cpu: 1
+            memory: 4Gi