Update iris example

Author: ospillinger

docs/deployments/request-handlers.md

@@ -44,7 +44,7 @@ def post_inference(prediction, metadata):
 ```python
 import numpy as np
 
-iris_labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
+labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
 
 def pre_inference(sample, metadata):
     # Convert a dictionary of features to a flattened in list in the order expected by the model
@@ -63,7 +63,7 @@ def post_inference(prediction, metadata):
     probabilites = prediction[0][0]
     predicted_class_id = int(np.argmax(probabilites))
     return {
-        "class_label": iris_labels[predicted_class_id],
+        "class_label": labels[predicted_class_id],
         "class_index": predicted_class_id,
         "probabilities": probabilites,
     }

examples/iris/cortex.yaml

@@ -4,12 +4,18 @@
 - kind: api
   name: tensorflow
   model: s3://cortex-examples/iris/tensorflow.zip
+  request_handler: handlers/tensorflow.py
 
 - kind: api
   name: pytorch
   model: s3://cortex-examples/iris/pytorch.onnx
   request_handler: handlers/pytorch.py
 
+- kind: api
+  name: keras
+  model: s3://cortex-examples/iris/keras.onnx
+  request_handler: handlers/keras.py
+
 - kind: api
   name: xgboost
   model: s3://cortex-examples/iris/xgboost.onnx
@@ -19,8 +25,3 @@
   name: sklearn
   model: s3://cortex-examples/iris/sklearn.onnx
   request_handler: handlers/sklearn.py
-
-- kind: api
-  name: keras
-  model: s3://cortex-examples/iris/keras.onnx
-  request_handler: handlers/keras.py

examples/iris/handlers/keras.py

@@ -1,6 +1,6 @@
 import numpy as np
 
-iris_labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
+labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
 
 
 def pre_inference(sample, metadata):
@@ -18,7 +18,7 @@ def post_inference(prediction, metadata):
     probabilites = prediction[0][0]
     predicted_class_id = int(np.argmax(probabilites))
     return {
-        "class_label": iris_labels[predicted_class_id],
+        "class_label": labels[predicted_class_id],
         "class_index": predicted_class_id,
         "probabilities": probabilites,
     }

examples/iris/handlers/pytorch.py

@@ -1,6 +1,6 @@
 import numpy as np
 
-iris_labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
+labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
 
 
 def pre_inference(sample, metadata):
@@ -17,7 +17,7 @@ def pre_inference(sample, metadata):
 def post_inference(prediction, metadata):
     predicted_class_id = int(np.argmax(prediction[0][0]))
     return {
-        "class_label": iris_labels[predicted_class_id],
+        "class_label": labels[predicted_class_id],
         "class_index": predicted_class_id,
         "probabilites": prediction[0][0],
     }

examples/iris/handlers/sklearn.py

@@ -7,7 +7,7 @@
 logger = get_logger()
 s3 = boto3.client("s3")
 
-iris_labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
+labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
 
 scalars_obj = s3.get_object(Bucket="cortex-examples", Key="iris/scalars.json")
 scalars = json.loads(scalars_obj["Body"].read().decode("utf-8"))
@@ -28,4 +28,4 @@ def pre_inference(sample, metadata):
 
 def post_inference(prediction, metadata):
     predicted_class_id = prediction[0][0]
-    return {"class_label": iris_labels[predicted_class_id], "class_index": predicted_class_id}
+    return {"class_label": labels[predicted_class_id], "class_index": predicted_class_id}

examples/iris/handlers/tensorflow.py

@@ -0,0 +1,6 @@
+labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
+
+
+def post_inference(prediction, metadata):
+    label_index = int(prediction["response"]["class_ids"][0])
+    return labels[label_index]

examples/iris/handlers/xgboost.py

@@ -1,6 +1,6 @@
 import numpy as np
 
-iris_labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
+labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
 
 
 def pre_inference(sample, metadata):
@@ -16,4 +16,4 @@ def pre_inference(sample, metadata):
 
 def post_inference(prediction, metadata):
     predicted_class_id = prediction[0][0]
-    return {"class_label": iris_labels[predicted_class_id], "class_index": predicted_class_id}
+    return {"class_label": labels[predicted_class_id], "class_index": predicted_class_id}

examples/iris/models/keras_model.py

@@ -20,7 +20,6 @@
 scores = model.evaluate(X_test, y_test)
 print("\n%s: %.2f%%" % (model.metrics_names[1], scores[1] * 100))
 
-# Convert to ONNX model format
 onnx_model = keras2onnx.convert_keras(model)
 with open("keras.onnx", "wb") as f:
     f.write(onnx_model.SerializeToString())

examples/iris/models/pytorch_model.py

@@ -56,7 +56,6 @@ def forward(self, X):
 
 print("prediction accuracy {}".format(accuracy_score(test_y.data, predict_y.data)))
 
-# Convert to ONNX model format
 placeholder = torch.randn(1, 4)
 torch.onnx.export(
     model, placeholder, "pytorch.onnx", input_names=["input"], output_names=["species"]

examples/iris/models/sklearn_model.py

@@ -14,15 +14,13 @@
 print("mean:", scaler.mean_)
 print("standard deviation:", np.sqrt(scaler.var_))
 
-
 X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.8, random_state=42)
 
 logreg_model = LogisticRegression(solver="lbfgs", multi_class="multinomial")
 logreg_model.fit(X_train, y_train)
 
 print("Test data accuracy: {:.2f}".format(logreg_model.score(X_test, y_test)))
 
-# Convert to ONNX model format
 onnx_model = convert_sklearn(logreg_model, initial_types=[("input", FloatTensorType([1, 4]))])
 with open("sklearn.onnx", "wb") as f:
     f.write(onnx_model.SerializeToString())

examples/iris/models/xgboost_model.py

@@ -1,10 +1,18 @@
-iris_labels = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]
+import xgboost as xgb
+from sklearn.datasets import load_iris
+from sklearn.model_selection import train_test_split
+from onnxmltools.convert import convert_xgboost
+from onnxconverter_common.data_types import FloatTensorType
 
+iris = load_iris()
+X, y = iris.data, iris.target
+X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.8, random_state=42)
 
-def post_inference(prediction, metadata):
-    predicted_class_id = prediction[0][0]
-    return {
-        "class_label": iris_labels[predicted_class_id],
-        "class_index": predicted_class_id,
-        "probabilities": prediction[1][0],
-    }
+xgb_model = xgb.XGBClassifier()
+xgb_model = xgb_model.fit(X_train, y_train)
+
+print("Test data accuracy of the xgb classifier is {:.2f}".format(xgb_model.score(X_test, y_test)))
+
+onnx_model = convert_xgboost(xgb_model, initial_types=[("input", FloatTensorType([1, 4]))])
+with open("xgboost.onnx", "wb") as f:
+    f.write(onnx_model.SerializeToString())