Model 빌드/배포 및 테스트
By Bys on March 4, 2025
MLOps
DecisionTreeClassifier를 활용한 Model 빌드의 첫 걸음. 이 단계는 이 동안 설치한 Jupyter notebook, MLflow, MiniO, Seldon 을 활용해 모델을 빌드/배포 하고 테스트 하는 시나리오를 작성한다.
Jupyterlab 에서 모델 빌드
Model mlflow
!pip install numpy
!pip install DecisionTreeClassifier
!pip install mlflow
!pip install boto3
import numpy as np
from sklearn.tree import DecisionTreeClassifier
import mlflow
import os
X = np.array([
[1, 1],
[1, 2],
[1, 3],
[1, 4],
[2, 1],
[2, 2],
[2, 3],
[2, 4],
[3, 1],
[3, 2],
[3, 3],
[3, 4],
[4, 1],
[4, 2],
[4, 3],
[4, 4],
[10,1],
[10,15]
])
y = np.array([0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3])
os.environ['MLFLOW_TRACKING_USERNAME'] = "admin"
os.environ['MLFLOW_TRACKING_PASSWORD'] = "admin"
os.environ['MLFLOW_S3_ENDPOINT_URL']='http://minio.minio.svc.cluster.local:9000'
os.environ['AWS_ACCESS_KEY_ID']='mlflow'
os.environ['AWS_SECRET_ACCESS_KEY']='miniomlflow'
HOST = "https://mlflow.bys.asia/"
EXPERIMENT_NAME = "mlflow-test-experiment"
# Connect to local MLflow tracking server
mlflow.set_tracking_uri(HOST)
# Set the experiment name through which you will label all your exerpiments runs
mlflow.set_experiment(EXPERIMENT_NAME)
# enable autologging for scikit
mlflow.sklearn.autolog()
model = DecisionTreeClassifier(max_depth=3, criterion='gini',min_samples_leaf = 1 ,min_samples_split = 2)
with mlflow.start_run() as run:
model.fit(X, y)
여기까지는 Jupyter notebook 에서 진행.
MLflow(MiniO) 에 저장된 Artifact 다운로드
Jupyter notebook 에서 모델 빌드를 하고 나서 저장된 Artifact 중 model.pkl, requirements.txt 파일을 다운로드 받는다.
1. Dockerfile
FROM python:3.11-bullseye
WORKDIR /microservice
RUN pip install --upgrade pip
COPY base_requirements.txt /microservice/
RUN pip install -r base_requirements.txt
COPY requirements.txt /microservice/
RUN pip install -r requirements.txt
COPY Predictor.py model.pkl /microservice/
CMD seldon-core-microservice $MODEL_NAME --service-type $SERVICE_TYPE --grpc-port ${GRPC_PORT} --metrics-port ${METRICS_PORT} --http-port ${HTTP_PORT}
2. Predictor.py
import joblib
class Predictor(object):
def __init__(self):
self.model = joblib.load('model.pkl')
def predict(self, data_array, column_names):
return self.model.predict_proba(data_array)
3. base_requirements.txt
seldon-core
joblib
4. model.pkl && requirements.txt
위 파일을 모두 한 폴더에 정리한 후 Dockerfile을 빌드한다.
Seldon
아래 SeldonDeployment 파일에서 이미지를 배포한 이미지로 수정하여 배포하면 이전 단계에서 학습시킨 model.pkl 을 파일을 배포할 수 있다.
seldondeployment.yaml
apiVersion: machinelearning.seldon.io/v1
kind: SeldonDeployment
metadata:
labels:
app: seldon
name: model-v3
spec:
annotations:
project_name: test
deployment_version: "v3"
# seldon.io/engine-separate-pod: "true"
name: test-specs
predictors:
- componentSpecs:
- spec:
containers:
- image: 202949997891.dkr.ecr.ap-northeast-2.amazonaws.com/common/build:mlv3
imagePullPolicy: Always
name: predictor
env:
- name: MODEL_NAME
value: "Predictor"
- name: SERVICE_TYPE
value: MODEL
- name: GRPC_PORT
value: "5007"
- name: METRICS_PORT
value: "6007"
- name: HTTP_PORT
value: "9000"
hpaSpec:
maxReplicas: 2
metrics:
- resource:
name: cpu
targetAverageUtilization: 80
type: Resource
minReplicas: 1
graph:
children:
name: predictor
type: MODEL
endpoint:
type: REST
service_host: localhost
service_port: 9000
# logger:
# url: http://logger/
# mode: all
name: predictor
annotations:
predictor_version: "v3"
team: opendatahub
seldon.io/svc-name: model-test
labels:
team: opendatahub
version: v1
replicas: 1
아래의 Ingress를 통해 seldon-operator를 통해 배포되는 서비스를 노출할 수 있다.
ingress.yaml
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
annotations:
project_name: model-test
deployment_version: v1
alb.ingress.kubernetes.io/group.name: mlops
alb.ingress.kubernetes.io/subnets: bys-dev-ue1-sbn-1a-extelb, bys-dev-ue1-sbn-1b-extelb, bys-dev-ue1-sbn-1c-extelb, bys-dev-ue1-sbn-1d-extelb, bys-dev-ue1-sbn-1f-extelb
alb.ingress.kubernetes.io/scheme : internet-facing
alb.ingress.kubernetes.io/listen-ports: '[{"HTTPS": 443}]'
alb.ingress.kubernetes.io/certificate-arn: arn:aws:acm:us-east-1:558846430793:certificate/a5207b24-ae67-49ac-b34e-f34ed0088bca
alb.ingress.kubernetes.io/security-groups: sg-07e6c272df0bed7ee
alb.ingress.kubernetes.io/healthcheck-path: /api/v1.0/predictions
alb.ingress.kubernetes.io/healthcheck-interval-seconds: '10'
alb.ingress.kubernetes.io/healthcheck-timeout-seconds: '5'
alb.ingress.kubernetes.io/healthy-threshold-count: '2'
alb.ingress.kubernetes.io/unhealthy-threshold-count: '2'
alb.ingress.kubernetes.io/healthcheck-port: traffic-port
alb.ingress.kubernetes.io/success-codes: 200-405
alb.ingress.kubernetes.io/target-type: ip
alb.ingress.kubernetes.io/tags: auto-delete=no
labels:
app: seldon
name: model-test
spec:
ingressClassName: "alb"
rules:
- host: model-test.bys.asia
http:
paths:
- path: /
pathType: Prefix
backend:
service:
name: model-test
port:
number: 8000
Test
X = np.array([
[1, 1],
[1, 2],
[1, 3],
[1, 4],
[2, 1],
[2, 2],
[2, 3],
[2, 4],
[3, 1],
[3, 2],
[3, 3],
[3, 4],
[4, 1],
[4, 2],
[4, 3],
[4, 4],
])
y = np.array([0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2])
curl -X POST 'https://model-test.bys.asia/api/v1.0/predictions' --header 'Content-Type: application/json' --data-raw '{ "data": { "ndarray": [[6,0]] }}'
{"data":{"names":["t:0","t:1","t:2"],"ndarray":[[0.0,0.0,1.0]]},"meta":{"requestPath":{"model-test-predictor":"202949997891.dkr.ecr.ap-northeast-2.amazonaws.com/common/build:mlv2"}}}
curl -X POST 'https://model-test.bys.asia/api/v1.0/predictions' --header 'Content-Type: application/json' --data-raw '{ "data": { "ndarray": [[2, 4]] }}'
{"data":{"names":["t:0","t:1","t:2"],"ndarray":[[0.0,1.0,0.0]]},"meta":{"requestPath":{"model-test-predictor":"202949997891.dkr.ecr.ap-northeast-2.amazonaws.com/common/build:mlv2"}}}
curl -X POST 'https://model-test.bys.asia/api/v1.0/predictions' --header 'Content-Type: application/json' --data-raw '{ "data": { "ndarray": [[10,0]] }}'
{"data":{"names":["t:0","t:1","t:2"],"ndarray":[[0.0,0.0,1.0]]},"meta":{"requestPath":{"predictor":"202949997891.dkr.ecr.ap-northeast-2.amazonaws.com/common/build:mlv2"}}}
curl -X POST 'https://model-test.bys.asia/api/v1.0/predictions' --header 'Content-Type: application/json' --data-raw '{ "data": { "ndarray": [[15,4]] }}'
{"data":{"names":["t:0","t:1","t:2"],"ndarray":[[0.0,0.0,1.0]]},"meta":{"requestPath":{"predictor":"202949997891.dkr.ecr.ap-northeast-2.amazonaws.com/common/build:mlv2"}}}
X = np.array([
[1, 1],
[1, 2],
[1, 3],
[1, 4],
[2, 1],
[2, 2],
[2, 3],
[2, 4],
[3, 1],
[3, 2],
[3, 3],
[3, 4],
[4, 1],
[4, 2],
[4, 3],
[4, 4],
[10,1],
[10,15]
])
y = np.array([0, 0, 1, 1, 0, 0, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3])
curl -X POST 'https://model-test.bys.asia/api/v1.0/predictions' --header 'Content-Type: application/json' --data-raw '{ "data": { "ndarray": [[6,0]] }}'
{"data":{"names":["t:0","t:1","t:2","t:3"],"ndarray":[[0.0,0.0,1.0,0.0]]},"meta":{"requestPath":{"predictor":"202949997891.dkr.ecr.ap-northeast-2.amazonaws.com/common/build:mlv3"}}}
curl -X POST 'https://model-test.bys.asia/api/v1.0/predictions' --header 'Content-Type: application/json' --data-raw '{ "data": { "ndarray": [[10,0]] }}'
{"data":{"names":["t:0","t:1","t:2","t:3"],"ndarray":[[0.0,0.0,0.0,1.0]]},"meta":{"requestPath":{"predictor":"202949997891.dkr.ecr.ap-northeast-2.amazonaws.com/common/build:mlv3"}}}
curl -X POST 'https://model-test.bys.asia/api/v1.0/predictions' --header 'Content-Type: application/json' --data-raw '{ "data": { "ndarray": [[15,4]] }}'
{"data":{"names":["t:0","t:1","t:2","t:3"],"ndarray":[[0.0,0.0,0.0,1.0]]},"meta":{"requestPath":{"predictor":"202949997891.dkr.ecr.ap-northeast-2.amazonaws.com/common/build:mlv3"}}}
분석
- X: 16개의 데이터 포인트, 각각 2개의 특성을 가짐
- y: 각 데이터 포인트의 클래스 레이블 (0, 1, 2 중 하나)
이 모델은 다음과 같은 패턴을 학습하게 된다.
첫 번째 특성이 1 또는 2인 경우:
대체로 클래스 0과 1로 분류됨
두 번째 특성이 1-2면 클래스 0, 3-4면 클래스 1인 패턴이 보임
첫 번째 특성이 3 또는 4인 경우:
모두 클래스 2로 분류됨
model v3 에서는 (10,1), (10, 15) 에 대한 데이터 클래스를 3으로 추가 제공하였고, 추론결과 (15,4)와 같은 데이터는 t4 클래스로 분류되는 것을 알 수 있다.
mlops