Extend Katib API for NAS jobs (#327)

* Add fields to studyjob structure

* Change nasjob yaml file

* Change parameter type

* Add Parameter Type=range

* Change API

* Change input size

* Reset API structure

* Change StudyJob API structure

* Remove Range parameter

* Fix api.proto

* Fix gopkg.toml

* Remove old nasjob file

* Fix nasjob.yaml

* Add custom suggestion

* Add blank NAS suggestion
Change Katib API to process yaml file for NAS

* Add correct YAML file for NAS example

* Fix newline

* Change StudyID to 1

* Add jobType parameter in Parsing

* Remove changes in manager

* Add NasConfig inside Yaml file

* Fix name in nasConfig

* Fix get StudyConfig in NAS

* Add JobType in all services

* Add job_type in bayesian_service

* Add pointers in NasConfig structure

* Fix Pointer in API

* Add consts for jobType
Remove return from populateCommonConfigFields

* Move const jobType to const file

* Remove Range parameter

* Modify YAML file for NAS jobs

* Add getStudyJobType function in GRPC server

* Add blank GetStudyJobType func in manager

* Fix metrics collector

* Remove jobType from getStudy

* Remove getStudyJobType from manager

* Add NAS RL yaml deployment

* Change worker to GPU

* Clean nasrl suggestion

* Add -u inside training-container

* Fix namespace in worker template

cmd/manager/main.go

@@ -91,8 +91,8 @@ func (s *server) GetTrials(ctx context.Context, in *api_pb.GetTrialsRequest) (*a
 }
 
 func (s *server) GetTrial(ctx context.Context, in *api_pb.GetTrialRequest) (*api_pb.GetTrialReply, error) {
-        t, err := dbIf.GetTrial(in.TrialId)
-        return &api_pb.GetTrialReply{Trial: t}, err
+	t, err := dbIf.GetTrial(in.TrialId)
+	return &api_pb.GetTrialReply{Trial: t}, err
 }
 
 func (s *server) GetSuggestions(ctx context.Context, in *api_pb.GetSuggestionsRequest) (*api_pb.GetSuggestionsReply, error) {

cmd/suggestion/nasrl/Dockerfile

@@ -0,0 +1,8 @@
+FROM python:3
+
+ADD . /usr/src/app/github.com/kubeflow/katib
+WORKDIR /usr/src/app/github.com/kubeflow/katib/cmd/suggestion/nasrl
+RUN pip install --no-cache-dir -r requirements.txt
+ENV PYTHONPATH /usr/src/app/github.com/kubeflow/katib:/usr/src/app/github.com/kubeflow/katib/pkg/api/python
+
+ENTRYPOINT ["python", "-u", "main.py"]

cmd/suggestion/nasrl/main.py

@@ -0,0 +1,29 @@
+import grpc
+from concurrent import futures
+
+import time
+
+from pkg.api.python import api_pb2_grpc
+from pkg.suggestion.nasrl_service import NasrlService
+from pkg.suggestion.types import DEFAULT_PORT
+from logging import getLogger, StreamHandler, INFO, DEBUG
+
+
+_ONE_DAY_IN_SECONDS = 60 * 60 * 24
+
+
+def serve():
+    print("NAS RL Suggestion Service")
+    server = grpc.server(futures.ThreadPoolExecutor(max_workers=10))
+    api_pb2_grpc.add_SuggestionServicer_to_server(NasrlService(), server)
+    server.add_insecure_port(DEFAULT_PORT)
+    print("Listening...")
+    server.start()
+    try:
+        while True:
+            time.sleep(_ONE_DAY_IN_SECONDS)
+    except KeyboardInterrupt:
+        server.stop(0)
+
+if __name__ == "__main__":
+    serve()

cmd/suggestion/nasrl/requirements.txt

@@ -0,0 +1,9 @@
+grpcio
+duecredit
+cloudpickle==0.5.6
+numpy>=1.13.3
+scikit-learn>=0.19.0
+scipy>=0.19.1
+forestci
+protobuf
+googleapis-common-protos

examples/nasjob-example-RL.yaml

@@ -0,0 +1,120 @@
+apiVersion: "kubeflow.org/v1alpha1"
+kind: StudyJob
+metadata:
+  namespace: kubeflow
+  labels:
+    controller-tools.k8s.io: "1.0"
+  name: nas-rl-example
+spec:
+  studyName: nas-rl-example
+  owner: crd
+  optimizationtype: maximize
+  objectivevaluename: Validation-Accuracy
+  optimizationgoal: 0.99
+  requestcount: 3
+  metricsnames:
+    - accuracy
+  nasConfig:
+    graphConfig:
+      numLayers: 8
+      inputSize:
+        - 32
+        - 32
+        - 3
+      outputSize:
+        - 10
+    operations:
+      - operationType: convolution
+        parameterconfigs:
+          - name: filter_size
+            parametertype: categorical
+            feasible:
+              list:
+              - "3"
+              - "5"
+              - "7"
+          - name: num_filter
+            parametertype: categorical
+            feasible:
+              list:
+              - "32"
+              - "48"
+              - "64"
+              - "96"
+              - "128"
+          - name: stride
+            parametertype: categorical
+            feasible:
+              list:
+              - "1"
+              - "2"
+      - operationType: reduction
+        parameterconfigs:
+          - name: reduction_type
+            parametertype: categorical
+            feasible:
+              list:
+              - max_pooling
+              - avg_pooling
+          - name: pool_size
+            parametertype: int
+            feasible:
+              min: "2"
+              max: "3"
+              step: "1"
+  workerSpec:
+    goTemplate:
+        rawTemplate: |-
+          apiVersion: batch/v1
+          kind: Job
+          metadata:
+            name: {{.WorkerID}}
+            namespace: {{.NameSpace}}
+          spec:
+            template:
+              spec:
+                containers:
+                - name: {{.WorkerID}}
+                  image: docker.io/deepermind/training-container-nas
+                  command:
+                  - "python3.5"
+                  - "-u"
+                  - "RunTrial.py"
+                  {{- with .HyperParameters}}
+                  {{- range .}}
+                  - "--{{.Name}}={{.Value}}"
+                  {{- end}}
+                  {{- end}}
+                  resources:
+                    limits:
+                      nvidia.com/gpu: 1
+                restartPolicy: Never
+  suggestionSpec:
+    suggestionAlgorithm: "nasrl"
+    suggestionParameters:
+      - name: "lstm_num_cells"
+        value: "64"
+      - name: "lstm_num_layers"
+        value: "1"
+      - name: "lstm_keep_prob"
+        value: "1.0"
+      - name: "optimizer"
+        value: "adam"
+      - name: "init_learning_rate"
+        value: "1e-3"
+      - name: "lr_decay_start"
+        value: "0"
+      - name: "lr_decay_every"
+        value: "1000"
+      - name: "lr_decay_rate"
+        value: "0.9"
+      - name: "skip-target"
+        value: "0.4"
+      - name: "skip-weight"
+        value: "0.8"
+      - name: "l2_reg"
+        value: "0"
+      - name: "entropy_weight"
+        value: "1e-4"
+      - name: "baseline_decay"
+        value: "0.9999"

manifests/vizier/suggestion/reinforcementlearning/deployment.yaml

@@ -0,0 +1,23 @@
+apiVersion: extensions/v1beta1
+kind: Deployment
+metadata:
+  name: vizier-suggestion-nasrl
+  namespace: kubeflow
+  labels:
+    app: vizier
+    component: suggestion-nasrl
+spec:
+  replicas: 1
+  template:
+    metadata:
+      name: vizier-suggestion-nasrl
+      labels:
+        app: vizier
+        component: suggestion-nasrl
+    spec:
+      containers:
+      - name: vizier-suggestion-nasrl
+        image: docker.io/deepermind/katib-nasrl-suggestion
+        ports:
+        - name: api
+          containerPort: 6789

manifests/vizier/suggestion/reinforcementlearning/service.yaml

@@ -0,0 +1,17 @@
+apiVersion: v1
+kind: Service
+metadata:
+  name: vizier-suggestion-nasrl
+  namespace: kubeflow
+  labels:
+    app: vizier
+    component: suggestion-nasrl
+spec:
+  type: ClusterIP
+  ports:
+    - port: 6789
+      protocol: TCP
+      name: api
+  selector:
+    app: vizier
+    component: suggestion-nasrl