如何从 gcp 存储桶中读取 keras.h5 model

Question

我有一个 keras model 作为 h5 文件保存在 gcp 存储桶存储中，我使用以下代码读取：

from keras.models import load_model
import h5py
import gcsfs

FS = gcsfs.GCSFileSystem(project="bucketname")
with FS.open(fn_model, 'rb') as model_file:
     model_gcs = h5py.File(model_file, 'r')
     myModel = load_model(model_gcs)

我收到错误AttributeError: 'str' object has no attribute 'decode'

然后我尝试了：

from keras.models import load_model
import h5py
import gcsfs

FS = gcsfs.GCSFileSystem(project="bucketname")
with FS.open(fn_model, 'rb', "utf-8") as model_file:
     model_gcs = h5py.File(model_file, 'r')
     myModel = load_model(model_gcs)

现在错误是unsupported operand type(s) for +: 'int' and 'str'

我在stackoverflow中尝试了某人的答案，如下所示：

from tensorflow.python.lib.io import file_io
model_file = file_io.FileIO(fn_model, mode='rb')

temp_model_location = 'temp_model.h5'
temp_model_file = open(temp_model_location, 'wb')
temp_model_file.write(model_file.read())
temp_model_file.close()
model_file.close()
model = load_model(temp_model_location)

仍然有错误： 'str' object has no attribute 'decode'

我使用 gsutil 将.h5 文件复制到本地驱动器，然后尝试 load_model，但仍然错误'str' object has no attribute 'decode'

从 gcp 存储桶读取.h5 Keras model 的正确方法是什么？

Answer 1

从存储桶加载 model 时，我遇到了同样的问题。这就是我所做的。

1. 首先，您将 model 复制到 VM 中的特定位置，假设您复制到/home/jupyter/project/位置
2. 从本地磁盘加载复制的 model 并删除复制的文件

---

import subprocess
import os

model_name = "model_file.h5"
subprocess.call(f"gsutil cp gs://bucket_name/model_path/{model_name} /home/jupyter/project/",shell=True)
myModel = load_model(f"/home/jupyter/project/{model_name}")
os.remove(f"/home/jupyter/project/{model_name}")