如何使用 AWS SageMaker Notebook Instance 部署预训练模型？

Question

我有一个预训练模型，我从 S3 存储桶加载到 AWS SageMaker Notebook 实例中，并在提供测试图像以从 S3 存储桶进行预测后，它会根据需要为我提供准确的结果。我想部署它，以便拥有一个可以进一步与 AWS Lambda 函数和 AWS API GateWay 集成的端点，以便我可以将模型与实时应用程序一起使用。 知道如何从 AWS Sagemaker Notebook Instance 部署模型并获取其端点吗？ .ipynb 文件中的代码如下供参考。

import boto3
import pandas as pd
import sagemaker
#from sagemaker import get_execution_role
from skimage.io import imread
from skimage.transform import resize
import numpy as np
from keras.models import load_model
import os
import time
import json
#role = get_execution_role()
role = sagemaker.get_execution_role()

bucketname = 'bucket' # bucket where the model is hosted
filename = 'test_model.h5' # name of the model
s3 = boto3.resource('s3')
image= s3.Bucket(bucketname).download_file(filename, 'test_model_new.h5')
model= 'test_model_new.h5'

model = load_model(model)

bucketname = 'bucket' # name of the bucket where the test image is hosted
filename = 'folder/image.png' # prefix
s3 = boto3.resource('s3')
file= s3.Bucket(bucketname).download_file(filename, 'image.png')
file_name='image.png'

test=np.array([resize(imread(file_name), (137, 310, 3))])

test_predict = model.predict(test)

print ((test_predict > 0.5).astype(np.int))

Answer 1

这是对我有用的解决方案。只需按照以下步骤操作即可。

1 - 在 SageMaker 的 jupyter 环境中加载您的模型

from keras.models import load_model

model = load_model (<Your Model name goes here>) #In my case it's model.h5

2 - 现在模型已加载，在

的帮助下将其转换为AWS 所需的protobuf format

def convert_h5_to_aws(loaded_model):

from tensorflow.python.saved_model import builder
from tensorflow.python.saved_model.signature_def_utils import predict_signature_def
from tensorflow.python.saved_model import tag_constants

model_version = '1'
export_dir = 'export/Servo/' + model_version
# Build the Protocol Buffer SavedModel at 'export_dir'
builder = builder.SavedModelBuilder(export_dir)
# Create prediction signature to be used by TensorFlow Serving Predict API
signature = predict_signature_def(
    inputs={"inputs": loaded_model.input}, outputs={"score": loaded_model.output})
from keras import backend as K

with K.get_session() as sess:
    # Save the meta graph and variables
    builder.add_meta_graph_and_variables(
        sess=sess, tags=[tag_constants.SERVING], signature_def_map={"serving_default": signature})
    builder.save()
import tarfile
with tarfile.open('model.tar.gz', mode='w:gz') as archive:
    archive.add('export', recursive=True)
import sagemaker

sagemaker_session = sagemaker.Session()
inputs = sagemaker_session.upload_data(path='model.tar.gz', key_prefix='model')
convert_h5_to_aws(model):

3 - 现在你可以在

的帮助下部署你的模型了

!touch train.py
from sagemaker.tensorflow.model import TensorFlowModel
sagemaker_model = TensorFlowModel(model_data = 's3://' + sagemaker_session.default_bucket() + '/model/model.tar.gz',
                                  role = role,
                                  framework_version = '1.15.2',
                                  entry_point = 'train.py')
%%timelog
predictor = sagemaker_model.deploy(initial_instance_count=1,
                                   instance_type='ml.m4.xlarge')

这将生成可以在 Amazon SageMaker 的推理部分中看到的端点，在该端点的帮助下，您现在可以从 jupyter notebook 以及 Web 和移动应用程序进行预测。 Liam 的 Youtube tutorial 和 Priya 的 AWS blog 对我帮助很大。