ML 명령어 - mnist clothes 예제¶

tensorflow의 옷 데이터로 학습을 진행하고, 개인 객체 저장소에 업로드하여 IRIS Discovery Service 적재, 서빙하는 시나리오입니다.

학습¶

10개의 옷 카테고리로 이루어진 7만장의 흑백 옷이미지를 다운로드, 전처리, 학습합니다.

여기 를 참조하였습니다.

tensorflow 2.1.0 패키지가 필요합니다.

pip install tensorflow==2.1.0

import tensorflow as tf
from tensorflow import keras

fashion_mnist = keras.datasets.fashion_mnist
(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()

# scale the values to 0.0 to 1.0
train_images = train_images / 255.0
test_images = test_images / 255.0

# reshape for feeding into the model
train_images = train_images.reshape(train_images.shape[0], 28, 28, 1)
test_images = test_images.reshape(test_images.shape[0], 28, 28, 1)

class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']

print('\ntrain_images.shape: {}, of {}'.format(train_images.shape, train_images.dtype))
print('test_images.shape: {}, of {}'.format(test_images.shape, test_images.dtype))
print('\ntrain_labels.shape: {}, of {}'.format(train_labels.shape, train_labels.dtype))
print('test_labels.shape: {}, of {}'.format(test_labels.shape, test_labels.dtype))

model = keras.Sequential([
    keras.layers.Conv2D(input_shape=(28,28,1), filters=8, kernel_size=3,
                        strides=2, activation='relu', name='Conv1'),
    keras.layers.Flatten(),
    keras.layers.Dense(10, activation=tf.nn.softmax, name='Softmax')
])
model.summary()

testing = False
epochs = 5

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])
model.fit(train_images, train_labels, epochs=epochs)

test_loss, test_acc = model.evaluate(test_images, test_labels)
print('\nTest accuracy: {}'.format(test_acc))

개인 객체저장소에 모델 업로드¶

IRIS Discovery Service에 적재하기 위해, tar 파일로 압축하여 개인 객체저장소에 업로드합니다.

boto3 패키지가 필요합니다.

pip install boto3

아래 인자를 입력해주세요.

bucket : 개인 객체 저장소의 bucket

key : 개인 객체 저장소의 key

endpoint_url : 개인 객체 저장소의 url

aws_access_key_id : 개인 객체 저장소의 access_key

aws_secret_access_key : 개인 객체 저장소의 secret_access_key

import tempfile
import tarfile
import os
import boto3
import tensorflow as tf

# 개인 객체저장소 설정
bucket = 'iris'
key = 'clothes/model.tar'
setting = {
    'endpoint_url': "http://192.168.102.138:9003",
    'verify': False,
    'aws_access_key_id': 'minio',
    'aws_secret_access_key': 'minio123'
}

# 모델 생성
export_path = tempfile.mkdtemp()
tf.keras.models.save_model(
    model,
    export_path,
    overwrite=True,
    include_optimizer=True,
    save_format=None,
    signatures=None,
    options=None
)

# 모델 압축
tar_name = export_path + '/model.tar'
with tarfile.open(tar_name, "w:tar") as tar:
    tar.add(export_path, arcname='./')

# 모델 업로드
cli = boto3.client('s3', **setting)
cli.upload_file(tar_name, bucket, key)

적재¶

IRIS Discovery Service에 모델을 적재합니다.

적재는 IRIS Discovery Service의 mlmodel import 를 사용합니다.

IRIS Discovery Service의 검색창에 아래 명령어를 입력합니다. path 옵션에 개인 객체저장소 정보, tar로 압축한 모델 경로를 입력합니다.

mlmodel import name=mnist_clothes type=tf category=classification algorithm=deep format=saved_model path=OBJECTSTORAGE.{CONNECTOR NAME}:clothes/model.tar

결과

result
ok

배포¶

IRIS Discovery Service가 관리하는 tensorflow serving에 모델을 배포합니다.

배포는 IRIS Discovery Service의 mlmodel deploy 를 사용합니다.

IRIS Discovery Service의 검색창에 아래 명령어를 입력합니다.

mlmodel deploy mnist_clothes label='first'

결과

mnist_clothes이 root_mnist_clothes 이름으로 배포되었습니다.

result	latest_version	serving_name
ok	1	root_mnist_clothes

서빙 상태 확인¶

배포한 mnist_clothes모델의 서빙 상태를 확인합니다.

서빙 상태 확인은 IRIS Discovery Service의 serving status 를 사용합니다.

IRIS Discovery Service의 검색창에 아래 명령어를 입력합니다.

serving status mnist_clothes

결과

mnist_clothes모델로 생성한 version 1이 사용 가능한 상태로 배포되었습니다.

version	state	label
1	AVAILABLE	first

예측¶

배포된 모델에 대해 4가지 유형의 예측 방법이 있습니다.

python 스크립트 방식
DSL 설정파일 방식
DSL 데이터 소스 입력 방식
curl 방식

이중 python 스크립트 방식, DSL 설정파일 방식, curl 방식에 대해 진행합니다. DSL 데이터 소스 입력 방식은 테스트 데이터가 IRIS에 업로드 되어있어야 합니다. 이에 관한 유즈케이스는 범용명령어 , mnist 숫자 모델 적재, 예측 을 참조해주세요.

python 스크립트 방식¶

시각화를 위해 matplotlib 패키지가 필요합니다.

pip install matplotlib

주의할점은 서빙에 직접 요청을 하기 때문에, url에 model_name(mnist_clothes) 대신 serving_name(root_mnist_clothes)을 입력해야 합니다.

import json, requests
import matplotlib.pyplot as plt
from tensorflow import keras
import numpy as np

url = 'http://192.168.100.180:8501/v1/models/root_mnist_clothes/versions/1:predict'

def show(idx, title):
    plt.figure()
    plt.imshow(test_images[idx].reshape(28,28))
    plt.axis('off')
    plt.title('\n\n{}'.format(title), fontdict={'size': 16})

def preprocessing():
    fashion_mnist = keras.datasets.fashion_mnist
    (train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
    # scale the values to 0.0 to 1.0
    test_images = test_images / 255.0
    # reshape for feeding into the model
    test_images = test_images.reshape(test_images.shape[0], 28, 28, 1)
    return test_images, test_labels

test_images, test_labels = preprocessing()
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
                   'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']

data = json.dumps({"signature_name": "serving_default", "instances": test_images[0:3].tolist()})
headers = {"content-type": "application/json"}
json_response = requests.post(url, data=data, headers=headers)
predictions = json.loads(json_response.text)['predictions']
for i in range(3):
    show(i, 'The model thought this was a {} (class {}), and it was actually a {} (class {})'.format(
        class_names[np.argmax(predictions[i])], test_labels[i], class_names[np.argmax(predictions[i])], test_labels[i]))

결과

DSL 설정파일 방식¶

앞서 python 스크립트 방식에서 서빙에 전달한 json 포맷을 개인 객체저장소에 업로드하고, DSL에 설정 경로를 입력하여 예측합니다.

개인 객체저장소에 설정을 업로드 합니다. 아래 인자를 입력해주세요.

bucket : 개인 객체 저장소의 bucket
key : 개인 객체 저장소의 key
endpoint_url : 개인 객체 저장소의 url
aws_access_key_id : 개인 객체 저장소의 access_key
aws_secret_access_key : 개인 객체 저장소의 secret_access_key

import json
import boto3
from tensorflow import keras

bucket = 'b-iris'
key = 'test/clothes.json'
setting = {
    'endpoint_url': "http://192.168.102.138:9003",
    'verify': False,
    'aws_access_key_id': 'minio',
    'aws_secret_access_key': 'minio123'
}

def preprocessing():
    fashion_mnist = keras.datasets.fashion_mnist
    (train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
    test_images = test_images / 255.0
    test_images = test_images.reshape(test_images.shape[0], 28, 28, 1)
    return test_images, test_labels

test_images, test_labels = preprocessing()

file_name = 'clothes.json'
contents = '{"signature_name": "serving_default", "instances": '+ str(test_images[0:3].tolist()) +'}'

with open(file_name, 'w') as f:
    json.dump(contents, f)

cli = boto3.client('s3', **setting)
cli.upload_file(file_name, bucket, key)

설정 파일의 경로를 입력하여 예측합니다.

예측(서빙)은 IRIS Discovery Service의 serving predict 를 사용합니다.

IRIS Discovery Service의 검색창에 아래 명령어를 입력합니다. conf 옵션에 개인 객체저장소의 CONNECTOR NAME을 입력합니다.

serving predict mnist_clothes1 conf=OBJECTSTORAGE.{CONNECTOR NAME}:test/clothes.json tag=(T-shirt/top, Trouser, Pullover, Dress, Coat, Sandal, Shirt, Sneaker, Bag, Ankle boot) version=1

결과

predictions	probability	interpreted
[0.0, 0.0, 0.0, 0.0, 0.0, 0.05, 0.0, 0.27, 0.0…	0.67	Ankle boot
[0.0, 0.0, 0.8, 0.0, 0.0, 0.0, 0.2, 0.0, 0.0, …	0.8	Pullover
[0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, …	1.0	Trouser

curl 방식¶

예측할 데이터를 x * 28 * 28 * 1 의 모양(shape)으로 가공하여, curl로 서빙에 요청합니다. x는 입력 데이터 개수.

curl -d '{"signature_name": "serving_default", "instances": [[[[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.011764705882352941], [0.00392156862745098], [0.0], [0.0], [0.027450980392156862], [0.0], [0.1450980392156863], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.00392156862745098], [0.00784313725490196], [0.0], [0.10588235294117647], [0.32941176470588235], [0.043137254901960784], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.4666666666666667], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.00392156862745098], [0.0], [0.0], [0.34509803921568627], [0.5607843137254902], [0.43137254901960786], [0.0], [0.0], [0.0], [0.0], [0.08627450980392157], [0.36470588235294116], [0.41568627450980394], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.01568627450980392], [0.0], [0.20784313725490197], [0.5058823529411764], [0.47058823529411764], [0.5764705882352941], [0.6862745098039216], [0.615686274509804], [0.6509803921568628], [0.5294117647058824], [0.6039215686274509], [0.6588235294117647], [0.5490196078431373], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.00784313725490196], [0.0], [0.043137254901960784], [0.5372549019607843], [0.5098039215686274], [0.5019607843137255], [0.6274509803921569], [0.6901960784313725], [0.6235294117647059], [0.6549019607843137], [0.6980392156862745], [0.5843137254901961], [0.592156862745098], [0.5647058823529412], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.00392156862745098], [0.0], [0.00784313725490196], [0.00392156862745098], [0.0], [0.011764705882352941], [0.0], [0.0], [0.45098039215686275], [0.4470588235294118], [0.41568627450980394], [0.5372549019607843], [0.6588235294117647], [0.6], [0.611764705882353], [0.6470588235294118], [0.6549019607843137], [0.5607843137254902], [0.615686274509804], [0.6196078431372549], [0.043137254901960784], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.00392156862745098], [0.0], [0.0], [0.0], [0.0], [0.0], [0.011764705882352941], [0.0], [0.0], [0.34901960784313724], [0.5450980392156862], [0.35294117647058826], [0.3686274509803922], [0.6], [0.5843137254901961], [0.5137254901960784], [0.592156862745098], [0.6627450980392157], [0.6745098039215687], [0.5607843137254902], [0.6235294117647059], [0.6627450980392157], [0.18823529411764706], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.00784313725490196], [0.01568627450980392], [0.00392156862745098], [0.0], [0.0], [0.0], [0.3843137254901961], [0.5333333333333333], [0.43137254901960786], [0.42745098039215684], [0.43137254901960786], [0.6352941176470588], [0.5294117647058824], [0.5647058823529412], [0.5843137254901961], [0.6235294117647059], [0.6549019607843137], [0.5647058823529412], [0.6196078431372549], [0.6627450980392157], [0.4666666666666667], [0.0]], [[0.0], [0.0], [0.00784313725490196], [0.00784313725490196], [0.00392156862745098], [0.00784313725490196], [0.0], [0.0], [0.0], [0.0], [0.10196078431372549], [0.4235294117647059], [0.4588235294117647], [0.38823529411764707], [0.43529411764705883], [0.4588235294117647], [0.5333333333333333], [0.611764705882353], [0.5254901960784314], [0.6039215686274509], [0.6039215686274509], [0.611764705882353], [0.6274509803921569], [0.5529411764705883], [0.5764705882352941], [0.611764705882353], [0.6980392156862745], [0.0]], [[0.011764705882352941], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.08235294117647059], [0.20784313725490197], [0.3607843137254902], [0.4588235294117647], [0.43529411764705883], [0.403921568627451], [0.45098039215686275], [0.5058823529411764], [0.5254901960784314], [0.5607843137254902], [0.6039215686274509], [0.6470588235294118], [0.6666666666666666], [0.6039215686274509], [0.592156862745098], [0.6039215686274509], [0.5607843137254902], [0.5411764705882353], [0.5882352941176471], [0.6470588235294118], [0.16862745098039217]], [[0.0], [0.0], [0.09019607843137255], [0.21176470588235294], [0.2549019607843137], [0.2980392156862745], [0.3333333333333333], [0.4627450980392157], [0.5019607843137255], [0.4823529411764706], [0.43529411764705883], [0.44313725490196076], [0.4627450980392157], [0.4980392156862745], [0.49019607843137253], [0.5450980392156862], [0.5215686274509804], [0.5333333333333333], [0.6274509803921569], [0.5490196078431373], [0.6078431372549019], [0.6313725490196078], [0.5647058823529412], [0.6078431372549019], [0.6745098039215687], [0.6313725490196078], [0.7411764705882353], [0.24313725490196078]], [[0.0], [0.26666666666666666], [0.3686274509803922], [0.35294117647058826], [0.43529411764705883], [0.4470588235294118], [0.43529411764705883], [0.4470588235294118], [0.45098039215686275], [0.4980392156862745], [0.5294117647058824], [0.5333333333333333], [0.5607843137254902], [0.49411764705882355], [0.4980392156862745], [0.592156862745098], [0.6039215686274509], [0.5607843137254902], [0.5803921568627451], [0.49019607843137253], [0.6352941176470588], [0.6352941176470588], [0.5647058823529412], [0.5411764705882353], [0.6], [0.6352941176470588], [0.7686274509803922], [0.22745098039215686]], [[0.27450980392156865], [0.6627450980392157], [0.5058823529411764], [0.40784313725490196], [0.3843137254901961], [0.39215686274509803], [0.3686274509803922], [0.3803921568627451], [0.3843137254901961], [0.4], [0.4235294117647059], [0.41568627450980394], [0.4666666666666667], [0.47058823529411764], [0.5058823529411764], [0.5843137254901961], [0.611764705882353], [0.6549019607843137], [0.7450980392156863], [0.7450980392156863], [0.7686274509803922], [0.7764705882352941], [0.7764705882352941], [0.7333333333333333], [0.7725490196078432], [0.7411764705882353], [0.7215686274509804], [0.1411764705882353]], [[0.06274509803921569], [0.49411764705882355], [0.6705882352941176], [0.7372549019607844], [0.7372549019607844], [0.7215686274509804], [0.6705882352941176], [0.6], [0.5294117647058824], [0.47058823529411764], [0.49411764705882355], [0.4980392156862745], [0.5725490196078431], [0.7254901960784313], [0.7647058823529411], [0.8196078431372549], [0.8156862745098039], [1.0], [0.8196078431372549], [0.6941176470588235], [0.9607843137254902], [0.9882352941176471], [0.984313725490196], [0.984313725490196], [0.9686274509803922], [0.8627450980392157], [0.807843137254902], [0.19215686274509805]], [[0.0], [0.0], [0.0], [0.047058823529411764], [0.2627450980392157], [0.41568627450980394], [0.6431372549019608], [0.7254901960784313], [0.7803921568627451], [0.8235294117647058], [0.8274509803921568], [0.8235294117647058], [0.8156862745098039], [0.7450980392156863], [0.5882352941176471], [0.3215686274509804], [0.03137254901960784], [0.0], [0.0], [0.0], [0.6980392156862745], [0.8156862745098039], [0.7372549019607844], [0.6862745098039216], [0.6352941176470588], [0.6196078431372549], [0.592156862745098], [0.043137254901960784]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]], [[0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]]]]}' -X POST http://192.168.100.180:8501/v1/models/mnist_clothes/versions/1:predict

결과

x * 10의 모양으로 예측 결과를 반환. x는 입력 데이터 개수, 10은 카테고리 개수.

{
    "predictions": [[1.34083416e-06, 6.62974609e-09, 4.16653876e-08, 6.56875301e-08, 1.06879767e-07, 0.00958874, 7.81568906e-06, 0.354254484, 0.00198251382, 0.634164929]
    ]
}