54 epsilon greedy

cluster-manager/pom.xml

@@ -4,7 +4,7 @@
   <groupId>io.seldon.clustermanager</groupId>
   <artifactId>seldon-cluster-manager</artifactId>
   <packaging>jar</packaging>
-  <version>0.1.4-SNAPSHOT</version>
+  <version>SNAPSHOT-MF-0</version>
   <name>seldon-cluster-manager</name>
   <url>http://maven.apache.org</url>
 

cluster-manager/src/main/java/io/seldon/clustermanager/k8s/SeldonDeploymentOperatorImpl.java

@@ -128,14 +128,15 @@ private V1.Container createEngineContainer(SeldonDeployment dep,PredictorSpec pr
 		// Add engine resources if specified
 		if (predictorDef.hasEngineResources())
 		    cBuilder.setResources(predictorDef.getEngineResources());
-		else // set default resource requests for cpu
-		    cBuilder.setResources(V1.ResourceRequirements.newBuilder().putRequests("cpu", Quantity.newBuilder().setString("0.1").build()));
-
+		else {// set default resource requests for cpu
+			final String DEFAULT_ENGINE_CPU_REQUEST = "0.1";
+		    cBuilder.setResources(V1.ResourceRequirements.newBuilder().putRequests("cpu", Quantity.newBuilder().setString(DEFAULT_ENGINE_CPU_REQUEST).build()));
+		}
 		return cBuilder.build();
 	}
 
 
-
+	;
 	private Set<String> getEnvNamesProto(List<EnvVar> envs)
 	{
 		Set<String> s = new HashSet<>();
@@ -184,7 +185,7 @@ private PredictiveUnit findPredictiveUnitForContainer(PredictiveUnit unit,String
 	    }
 	}
 
-	private V1.Container updateContainer(V1.Container c,PredictiveUnit pu,int idx)
+	private V1.Container updateContainer(V1.Container c,PredictiveUnit pu,int idx,String deploymentName,String predictorName)
 	{
 		V1.Container.Builder c2Builder = V1.Container.newBuilder(c);
 
@@ -257,6 +258,17 @@ private V1.Container updateContainer(V1.Container c,PredictiveUnit pu,int idx)
 		if (!envNames.contains(ENV_PREDICTIVE_UNIT_PARAMETERS))
 		    c2Builder.addEnv(EnvVar.newBuilder().setName(ENV_PREDICTIVE_UNIT_PARAMETERS).setValue(extractPredictiveUnitParametersAsJson(pu)));
 
+		//Add environment variable for the predictive unit ID, the predictor ID and the Deployment ID
+		final String ENV_PREDICTIVE_UNIT_ID = "PREDICTIVE_UNIT_ID";
+		final String ENV_PREDICTOR_ID = "PREDICTOR_ID";
+		final String ENV_SELDON_DEPLOYMENT_ID = "SELDON_DEPLOYMENT_ID";
+		if (!envNames.contains(ENV_PREDICTIVE_UNIT_ID))
+			c2Builder.addEnv(EnvVar.newBuilder().setName(ENV_PREDICTIVE_UNIT_ID).setValue(c.getName()));
+		if (!envNames.contains(ENV_PREDICTOR_ID))
+			c2Builder.addEnv(EnvVar.newBuilder().setName(ENV_PREDICTOR_ID).setValue(predictorName));
+		if (!envNames.contains(ENV_SELDON_DEPLOYMENT_ID))
+			c2Builder.addEnv(EnvVar.newBuilder().setName(ENV_SELDON_DEPLOYMENT_ID).setValue(deploymentName));
+
 		// Add a default lifecycle pre-stop if non exists
 		if (!c.hasLifecycle())
 		{
@@ -301,16 +313,19 @@ public SeldonDeployment defaulting(SeldonDeployment mlDep) {
 		SeldonDeployment.Builder mlBuilder = SeldonDeployment.newBuilder(mlDep);
 		int idx = 0;
 		String serviceName = mlDep.getSpec().getName();
+		String deploymentName = mlDep.getMetadata().getName();
+
 		for(PredictorSpec p : mlDep.getSpec().getPredictorsList())
 		{
 			ObjectMeta.Builder metaBuilder = ObjectMeta.newBuilder(p.getComponentSpec().getMetadata())
 				.putLabels(LABEL_SELDON_APP, serviceName);
 			mlBuilder.getSpecBuilder().getPredictorsBuilder(idx).getComponentSpecBuilder().setMetadata(metaBuilder);
 			int cIdx = 0;
 			mlBuilder.getSpecBuilder().getPredictorsBuilder(idx).getComponentSpecBuilder().getSpecBuilder().clearContainers();
+			String predictorName = p.getName();
 			for(V1.Container c : p.getComponentSpec().getSpec().getContainersList())
 			{
-				V1.Container c2 = this.updateContainer(c, findPredictiveUnitForContainer(mlDep.getSpec().getPredictors(idx).getGraph(),c.getName()),cIdx);
+				V1.Container c2 = this.updateContainer(c, findPredictiveUnitForContainer(mlDep.getSpec().getPredictors(idx).getGraph(),c.getName()),cIdx,deploymentName,predictorName);
 				mlBuilder.getSpecBuilder().getPredictorsBuilder(idx).getComponentSpecBuilder().getSpecBuilder().addContainers(cIdx, c2);	
 				updatePredictiveUnitBuilderByName(mlBuilder.getSpecBuilder().getPredictorsBuilder(idx).getGraphBuilder(),c2);
 				cIdx++;

cluster-manager/src/test/java/io/seldon/clustermanager/k8s/SeldonDeploymentDefaultingTest.java

@@ -38,7 +38,7 @@ public void testDefaulting() throws IOException
         Assert.assertTrue(mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).hasLivenessProbe());
         Assert.assertTrue(mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).hasReadinessProbe());
         Assert.assertTrue(mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).hasLifecycle());
-        Assert.assertEquals(2,mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).getEnvCount());
+        Assert.assertEquals("Incorrect number of environment variables in container",5,mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).getEnvCount());
         Assert.assertEquals(1,mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).getPortsCount());
         Assert.assertEquals("http",mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).getPorts(0).getName());
         Assert.assertEquals(Endpoint.EndpointType.REST_VALUE,mlDep2.getSpec().getPredictors(0).getGraph().getEndpoint().getType().getNumber());
@@ -56,7 +56,7 @@ public void testDefaultingGrpc() throws IOException
         Assert.assertTrue(mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).hasLivenessProbe());
         Assert.assertTrue(mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).hasReadinessProbe());
         Assert.assertTrue(mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).hasLifecycle());
-        Assert.assertEquals(2,mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).getEnvCount());
+        Assert.assertEquals("Incorrect number of environment variables in container",5,mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).getEnvCount());
         Assert.assertEquals(1,mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).getPortsCount());
         Assert.assertEquals("grpc",mlDep2.getSpec().getPredictors(0).getComponentSpec().getSpec().getContainers(0).getPorts(0).getName());
         Assert.assertEquals(Endpoint.EndpointType.GRPC_VALUE,mlDep2.getSpec().getPredictors(0).getGraph().getEndpoint().getType().getNumber());

docs/getting_started/minikube.md

@@ -72,7 +72,7 @@ In this session, we show how to wrap the sklearn iris classifier in the [seldon-
 
 3. Wrap your saved model using the core-python-wrapper docker image:
     ```bash
-    docker run -v $(pwd):/model seldonio/core-python-wrapper:0.5 /model IrisClassifier 0.1 seldonio --force
+    docker run -v $(pwd):/model seldonio/core-python-wrapper:0.6 /model IrisClassifier 0.1 seldonio --force
     ```
 
 4. Build the docker image locally

docs/wrappers/h2o.md

@@ -44,7 +44,7 @@ Detailed steps:
 3. Run the python wrapping scripts, with the additional ````--base-image``` argument:
 
 	```bash
-	docker run -v /path/to/your/model/folder:/model seldonio/core-python-wrapper:0.5 /model H2OModel 0.1 myrepo --base-image=H2OBase:1.0
+	docker run -v /path/to/your/model/folder:/model seldonio/core-python-wrapper:0.6 /model H2OModel 0.1 myrepo --base-image=H2OBase:1.0
 	```
 
 	"0.1" is the version of the docker image that will be created. "myrepo" is the name of your dockerhub repository.
@@ -88,7 +88,7 @@ Here we give a step by step example in which we will train and save a [H2O model
 
     ```bash
     cd ../../
-	docker run -v models/h2o_example:my_model seldonio/core-python-wrapper:0.5 my_model H2OModel 0.1 myrepo --base-image=H2OBase:1.0
+	docker run -v models/h2o_example:my_model seldonio/core-python-wrapper:0.6 my_model H2OModel 0.1 myrepo --base-image=H2OBase:1.0
     ``` 
 
     This will create a docker image "seldonio/h2omodel:0.1", which is ready to be deployed in seldon-core.

docs/wrappers/python.md

@@ -61,13 +61,13 @@ After you have copied the required files in your model folder, you run the Seldo
 In order to make things even simpler (and because we love Docker!) we have dockerised the wrapper script so that you don't need to install anything on your machine to run it - except Docker.
 
 ```
-docker run -v /path/to/model/dir:/my_model seldonio/core-python-wrapper:0.5 /my_model MnistClassifier 0.1 seldonio
+docker run -v /path/to/model/dir:/my_model seldonio/core-python-wrapper:0.6 /my_model MnistClassifier 0.1 seldonio
 ```
 
 Let's explain each piece of this command in more details.
 
 
-``` docker run seldonio/core-python-wrapper:0.5 ``` : run the core-python-wrapper container (version 0.5)
+``` docker run seldonio/core-python-wrapper:0.6 ``` : run the core-python-wrapper container (version 0.6)
 
 ``` -v /path/to/model/dir:/my_model ``` : Tells docker to mount your local folder to /my_model in the container. This is used to access your files and generate the wrapped model files. 
 
@@ -76,7 +76,7 @@ Let's explain each piece of this command in more details.
 For reference, here is the complete list of arguments that can be passed to the script.
 
 ```
-docker run -v /path:<model_path> seldonio/core-python-wrapper:0.5 
+docker run -v /path:<model_path> seldonio/core-python-wrapper:0.6 
 	<model_path>
 	<model_name>
 	<image_version>
@@ -108,7 +108,7 @@ Optional:
 Note that you can access the command line help of the script by using the -h or --help argument as follows:
 
 ```
-docker run seldonio/core-python-wrapper:0.5 -h
+docker run seldonio/core-python-wrapper:0.6 -h
 ```
 
 Note also that you could use the python script directly if you feel so enclined, but you would have to check out seldon-core and install some python libraries on your local machine - by using the docker image you don't have to care about these dependencies.

examples/routers/epsilon_greedy/EpsilonGreedy.py

@@ -0,0 +1,62 @@
+import random
+import numpy as np
+
+__version__ = "v1.1"
+
+def n_success_failures(features,reward):
+    n_predictions = features.shape[0]
+    n_success = int(reward*n_predictions)
+    n_failures = n_predictions - n_success
+    return n_success, n_failures
+
+class EpsilonGreedy(object):
+
+    def __init__(self,n_branches=None,epsilon=0.1,verbose=False):
+        print "Starting Epsilon Greedy Microservice, version {}".format(__version__)
+        if n_branches is None:
+            raise Exception("n_branches parameter must be given")
+        self.verbose = verbose
+        self.epsilon = epsilon
+        self.best_branch = 0
+        self.branches_success = [0 for _ in range(n_branches)]
+        self.branches_tries = [0 for _ in range(n_branches)]
+        self.n_branches = n_branches
+        if self.verbose:
+            print "Router initialised"
+            print "# branches:",self.n_branches
+            print "Epsilon:",self.epsilon
+            print
+
+    def route(self,features,feature_names):
+        x = random.random()
+        best_branch = self.best_branch
+        other_branches = [i for i in range(self.n_branches) if i!=best_branch]
+        selected_branch = best_branch if x>self.epsilon else random.choice(other_branches)
+        if self.verbose:
+            print "Routing"
+            print "Current best branch:",best_branch
+            print "Selected branch:",selected_branch
+            print
+        return selected_branch
+
+    def send_feedback(self,features,feature_names,routing,reward,truth):
+        if self.verbose:
+            print "Training"
+            print "Prev success #", self.branches_success
+            print "Prev tries #", self.branches_tries
+            print "Prev best branch:", self.best_branch
+        n_success, n_failures = n_success_failures(features,reward)
+        self.branches_success[routing] += n_success
+        self.branches_tries[routing] += n_success + n_failures
+        perfs = [
+            (self.branches_success[i]+1)/float(self.branches_tries[i]+1)
+            for i
+            in range(self.n_branches)
+        ]
+        self.best_branch = np.argmax(perfs)
+        if self.verbose:
+            print "New success #", self.branches_success
+            print "New tries #", self.branches_tries
+            print "New best branch:",self.best_branch
+            print
+

examples/routers/epsilon_greedy/requirements.txt

@@ -0,0 +1 @@
+numpy==1.11.2

helm-charts/seldon-core/values.yaml

@@ -7,7 +7,7 @@ cluster_manager:
   enabled: true
   image:
     pull_policy: IfNotPresent
-    tag: 0.1.4-SNAPSHOT
+    tag: SNAPSHOT-MF-0
   java_opts: ''
   rbac: false
   spring_opts: ''