Remove sample and prediction key

cli/cmd/get.go

@@ -289,7 +289,7 @@ func describeAPI(name string, resourcesRes *schema.GetResourcesResponse, flagVer
 
 	out := "\n" + console.Bold("url:  ") + apiEndpoint + "\n"
 	out += "\n" + console.Bold("debug url:  ") + apiEndpoint + "?debug=true" + "\n\n"
-	out += fmt.Sprintf("%s curl -X POST -H \"Content-Type: application/json\" %s -d @samples.json\n", console.Bold("curl:"), apiEndpoint)
+	out += fmt.Sprintf("%s curl -X POST -H \"Content-Type: application/json\" %s -d @sample.json\n", console.Bold("curl:"), apiEndpoint)
 	out += fmt.Sprintf(console.Bold("updated at:")+" %s\n\n", libtime.LocalTimestamp(updatedAt))
 
 	statusTable := table.Table{

cli/cmd/predict.go

@@ -39,18 +39,14 @@ func init() {
 	predictCmd.Flags().BoolVar(&predictDebug, "debug", false, "Predict with debug mode")
 }
 
-type PredictResponse struct {
-	Predictions []interface{} `json:"predictions"`
-}
-
 var predictCmd = &cobra.Command{
-	Use:   "predict API_NAME SAMPLES_FILE",
+	Use:   "predict API_NAME SAMPLE_FILE",
 	Short: "make predictions",
 	Long:  "Make predictions.",
 	Args:  cobra.ExactArgs(2),
 	Run: func(cmd *cobra.Command, args []string) {
 		apiName := args[0]
-		samplesJSONPath := args[1]
+		sampleJSONPath := args[1]
 
 		resourcesRes, err := getResourcesResponse()
 		if err != nil {
@@ -86,7 +82,7 @@ var predictCmd = &cobra.Command{
 		if predictDebug {
 			apiURL += "?debug=true"
 		}
-		predictResponse, err := makePredictRequest(apiURL, samplesJSONPath)
+		predictResponse, err := makePredictRequest(apiURL, sampleJSONPath)
 		if err != nil {
 			if strings.Contains(err.Error(), "503 Service Temporarily Unavailable") || strings.Contains(err.Error(), "502 Bad Gateway") {
 				errors.Exit(ErrorAPINotReady(apiName, resource.StatusCreating.Message()))
@@ -102,12 +98,12 @@ var predictCmd = &cobra.Command{
 	},
 }
 
-func makePredictRequest(apiURL string, samplesJSONPath string) (*PredictResponse, error) {
-	samplesBytes, err := files.ReadFileBytes(samplesJSONPath)
+func makePredictRequest(apiURL string, sampleJSONPath string) (interface{}, error) {
+	sampleBytes, err := files.ReadFileBytes(sampleJSONPath)
 	if err != nil {
 		errors.Exit(err)
 	}
-	payload := bytes.NewBuffer(samplesBytes)
+	payload := bytes.NewBuffer(sampleBytes)
 	req, err := http.NewRequest("POST", apiURL, payload)
 	if err != nil {
 		return nil, errors.Wrap(err, errStrCantMakeRequest)
@@ -119,11 +115,11 @@ func makePredictRequest(apiURL string, samplesJSONPath string) (*PredictResponse
 		return nil, err
 	}
 
-	var predictResponse PredictResponse
+	var predictResponse interface{}
 	err = json.DecodeWithNumber(httpResponse, &predictResponse)
 	if err != nil {
 		return nil, errors.Wrap(err, "prediction response")
 	}
 
-	return &predictResponse, nil
+	return predictResponse, nil
 }

dev/load.sh

@@ -19,11 +19,7 @@ SLEEP="0"  # seconds
 
 # Iris
 URL="https://a5a5afc3ba1bb11e9a4d20ab58369843-145237363.us-west-2.elb.amazonaws.com/iris/tensorflow"
-DATA='{ "samples": [ { "sepal_length": 5.2, "sepal_width": 3.6, "petal_length": 1.4, "petal_width": 0.3 } ] }'
-
-# Insurance
-# URL="https://ac47cdf2e99cb11e9ade10693263c378-1155928797.us-west-2.elb.amazonaws.com/insurance/cost"
-# DATA='{ "samples": [ { "age": 22, "bmi": 25, "children": 0, "region": "northeast", "sex": "female", "smoker": "no" } ] }'
+DATA='{ "sepal_length": 5.2, "sepal_width": 3.6, "petal_length": 1.4, "petal_width": 0.3 }'
 
 trap ctrl_c INT
 function ctrl_c() {

docs/cluster/cli.md

@@ -71,17 +71,17 @@ The `refresh` command behaves similarly to the `deploy` command. The key differe
 ## predict
 
 ```text
-Make predictions.
 Usage:
-  cortex predict API_NAME SAMPLES_FILE [flags]
+  cortex predict API_NAME SAMPLE_FILE [flags]
+
 Flags:
+      --debug               Predict with debug mode
   -d, --deployment string   deployment name
   -e, --env string          environment (default "dev")
   -h, --help                help for predict
-  -j, --json                print the raw json response
 ```
 
-The `predict` command converts samples from a JSON file into prediction requests and displays the response. This command is useful for quickly testing predictions.
+The `predict` command converts a sample from a JSON file into a prediction request and displays the response. This command is useful for quickly testing predictions.
 
 ## delete
 

docs/cluster/install.md

@@ -62,7 +62,7 @@ cortex get tensorflow
 
 # Classify a sample
 curl -X POST -H "Content-Type: application/json" \
-     -d '{ "samples": [ { "sepal_length": 5.2, "sepal_width": 3.6, "petal_length": 1.4, "petal_width": 0.3 } ] }' \
+     -d '{ "sepal_length": 5.2, "sepal_width": 3.6, "petal_length": 1.4, "petal_width": 0.3 }' \
      <API endpoint>
 ```
 

examples/image-classifier/sample_image_url.json

@@ -0,0 +1,3 @@
+{
+  "url": "https://bowwowinsurance.com.au/wp-content/uploads/2018/10/akita-700x700.jpg"
+}

examples/iris-classifier/README.md

@@ -61,7 +61,7 @@ url: http://***.amazonaws.com/iris/classifier
 
 $ curl http://***.amazonaws.com/iris/classifier \
     -X POST -H "Content-Type: application/json" \
-    -d '{ "samples": [{"sepal_length": 5.2, "sepal_width": 3.6, "petal_length": 1.4, "petal_width": 0.3}]}'
+    -d '{"sepal_length": 5.2, "sepal_width": 3.6, "petal_length": 1.4, "petal_width": 0.3}'
 
 iris-setosa
 ```

examples/iris-classifier/sample.json

@@ -0,0 +1,6 @@
+{
+  "sepal_length": 5.2,
+  "sepal_width": 3.6,
+  "petal_length": 1.4,
+  "petal_width": 0.3
+}

examples/sentiment-analysis/README.md

@@ -77,7 +77,7 @@ url: http://***.amazonaws.com/sentiment/analysis
 
 $ curl http://***.amazonaws.com/sentiment/analysis \
     -X POST -H "Content-Type: application/json" \
-    -d '{ "samples": [{"input": "The movie was great!"}]}'
+    -d '{"input": "The movie was great!"}'
 
 positive
 ```

examples/sentiment-analysis/sample.json

@@ -0,0 +1,3 @@
+{
+  "input": "The movie was great!"
+}

examples/text-generator/README.md

@@ -64,7 +64,7 @@ The output above indicates that one replica of the API was requested and one rep
 ```bash
 $ curl http://***.amazonaws.com/text/generator \
     -X POST -H "Content-Type: application/json" \
-    -d '{"samples": [{"text": "machine learning"}]}'
+    -d '{"text": "machine learning"}'
 
 Machine learning, with more than one thousand researchers around the world today, are looking to create computer-driven machine learning algorithms that can also be applied to human and social problems, such as education, health care, employment, medicine, politics, or the environment...
 ```

examples/text-generator/sample.json

@@ -0,0 +1,3 @@
+{
+  "text": "machine learning"
+}

pkg/workloads/cortex/lib/api_utils.py

@@ -62,100 +62,81 @@ def status_code_metric(dimensions, status_code):
     return [{"MetricName": "StatusCode", "Dimensions": status_code_dimensions, "Value": 1}]
 
 
-def predictions_per_request_metric(dimensions, prediction_count):
-    return [
-        {"MetricName": "PredictionsPerRequest", "Dimensions": dimensions, "Value": prediction_count}
-    ]
-
-
-def extract_predicted_values(api, predictions):
-    predicted_values = []
-
+def extract_prediction(api, prediction):
     tracker = api.get("tracker")
-    for prediction in predictions:
-        if tracker.get("key") is not None:
-            key = tracker["key"]
-            if type(prediction) != dict:
-                raise ValueError(
-                    "failed to track key '{}': expected prediction to be of type dict but found '{}'".format(
-                        key, type(prediction)
-                    )
-                )
-            if prediction.get(key) is None:
-                raise ValueError(
-                    "failed to track key '{}': not found in prediction".format(tracker["key"])
+    if tracker.get("key") is not None:
+        key = tracker["key"]
+        if type(prediction) != dict:
+            raise ValueError(
+                "failed to track key '{}': expected prediction to be of type dict but found '{}'".format(
+                    key, type(prediction)
                 )
-            predicted_value = prediction[key]
-        else:
-            predicted_value = prediction
-
-        if tracker["model_type"] == "classification":
-            if type(predicted_value) != str and type(predicted_value) != int:
-                raise ValueError(
-                    "failed to track classification prediction: expected type 'str' or 'int' but encountered '{}'".format(
-                        type(predicted_value)
-                    )
+            )
+        if prediction.get(key) is None:
+            raise ValueError(
+                "failed to track key '{}': not found in prediction".format(tracker["key"])
+            )
+        predicted_value = prediction[key]
+    else:
+        predicted_value = prediction
+
+    if tracker["model_type"] == "classification":
+        if type(predicted_value) != str and type(predicted_value) != int:
+            raise ValueError(
+                "failed to track classification prediction: expected type 'str' or 'int' but encountered '{}'".format(
+                    type(predicted_value)
                 )
-        else:
-            if type(predicted_value) != float and type(predicted_value) != int:  # allow ints
-                raise ValueError(
-                    "failed to track regression prediction: expected type 'float' or 'int' but encountered '{}'".format(
-                        type(predicted_value)
-                    )
+            )
+    else:
+        if type(predicted_value) != float and type(predicted_value) != int:  # allow ints
+            raise ValueError(
+                "failed to track regression prediction: expected type 'float' or 'int' but encountered '{}'".format(
+                    type(predicted_value)
                 )
-        predicted_values.append(predicted_value)
+            )
+    return predicted_value
 
-    return predicted_values
 
-
-def prediction_metrics(dimensions, api, predicted_values):
+def prediction_metrics(dimensions, api, prediction):
     metric_list = []
     tracker = api.get("tracker")
-    for predicted_value in predicted_values:
-        if tracker["model_type"] == "classification":
-            dimensions_with_class = dimensions + [{"Name": "Class", "Value": str(predicted_value)}]
-            metric = {
-                "MetricName": "Prediction",
-                "Dimensions": dimensions_with_class,
-                "Unit": "Count",
-                "Value": 1,
-            }
-
-            metric_list.append(metric)
-        else:
-            metric = {
-                "MetricName": "Prediction",
-                "Dimensions": dimensions,
-                "Value": float(predicted_value),
-            }
-            metric_list.append(metric)
+    if tracker["model_type"] == "classification":
+        dimensions_with_class = dimensions + [{"Name": "Class", "Value": str(prediction)}]
+        metric = {
+            "MetricName": "Prediction",
+            "Dimensions": dimensions_with_class,
+            "Unit": "Count",
+            "Value": 1,
+        }
+
+        metric_list.append(metric)
+    else:
+        metric = {"MetricName": "Prediction", "Dimensions": dimensions, "Value": float(prediction)}
+        metric_list.append(metric)
     return metric_list
 
 
-def cache_classes(ctx, api, predicted_values, class_set):
-    for predicted_value in predicted_values:
-        if predicted_value not in class_set:
-            upload_class(ctx, api["name"], predicted_value)
-            class_set.add(predicted_value)
+def cache_classes(ctx, api, prediction, class_set):
+    if prediction not in class_set:
+        upload_class(ctx, api["name"], prediction)
+        class_set.add(prediction)
 
 
-def post_request_metrics(ctx, api, response, predictions, class_set):
+def post_request_metrics(ctx, api, response, prediction_payload, class_set):
     api_name = api["name"]
     api_dimensions = api_metric_dimensions(ctx, api_name)
     metrics_list = []
     metrics_list += status_code_metric(api_dimensions, response.status_code)
 
-    if predictions is not None:
-        metrics_list += predictions_per_request_metric(api_dimensions, len(predictions))
-
+    if prediction_payload is not None:
         if api.get("tracker") is not None:
             try:
-                predicted_values = extract_predicted_values(api, predictions)
+                prediction = extract_prediction(api, prediction_payload)
 
                 if api["tracker"]["model_type"] == "classification":
-                    cache_classes(ctx, api, predicted_values, class_set)
+                    cache_classes(ctx, api, prediction, class_set)
 
-                metrics_list += prediction_metrics(api_dimensions, api, predicted_values)
+                metrics_list += prediction_metrics(api_dimensions, api, prediction)
             except Exception as e:
                 logger.warn(str(e), exc_info=True)