【问题标题】:Could not see hdf5 files while monitoring and saving keras modeli using ModelCheckPoint使用 ModelCheckPoint 监视和保存 keras modeli 时看不到 hdf5 文件
【发布时间】:2021-01-31 22:24:15
【问题描述】:

我正在运行 keras 模型并指示我想在给定目录中保存我的最佳模型的代码,因此 google colab 中的目录名称是

path ='/content/drive/My Drive/Colab Notebooks/BTU_Training2020/Weights/weights-improvement-{epoch:02d}-{val_accuracy:.2f}.hdf5'

我在 keras 模型的末尾编写了以下代码来保存和保存最佳模型:

checkpoint =ModelCheckpoint(filepath=path,monitor='val_accuracy',verbose=1,save_best_only=True,mode='max')
callbacks_list = [checkpoint]
model.fit(X,y,validation_split=0.33,epochs=150,batch_size=10,callbacks=callbacks_list,verbose=0)

它运行良好,并向我展示了一些这样的 cmets

Epoch 00001: val_accuracy improved from -inf to 0.55512, saving model to /content/drive/My Drive/Colab Notebooks/BTU_Training2020/Weights/weights-improvement-01-0.56.hdf5

Epoch 00002: val_accuracy did not improve from 0.55512

Epoch 00003: val_accuracy did not improve from 0.55512

Epoch 00004: val_accuracy did not improve from 0.55512

Epoch 00005: val_accuracy improved from 0.55512 to 0.57087, saving model to /content/drive/My Drive/Colab Notebooks/BTU_Training2020/Weights/weights-improvement-05-0.57.hdf5

Epoch 00006: val_accuracy did not improve from 0.57087

Epoch 00007: val_accuracy improved from 0.57087 to 0.62598, saving model to /content/drive/My Drive/Colab Notebooks/BTU_Training2020/Weights/weights-improvement-07-0.63.hdf5

Epoch 00008: val_accuracy did not improve from 0.62598

但是当我看到目录时,没有重量文件夹,每个文件夹都没有文件,你能解释一下是什么原因吗?

【问题讨论】:

    标签: python keras google-colaboratory


    【解决方案1】:

    您需要在文件路径中添加一个 ./。

    这是假设您尝试保存的文件夹已经创建。

    path ='./content/drive/My Drive/Colab Notebooks/BTU_Training2020/Weights/weights-improvement-{epoch:02d}-{val_accuracy:.2f}.hdf5'
    

    【讨论】:

    • 仍然没有效果,也许几分钟后结果就会清楚
    • 试试这个路径 ='./weights-improvement-{epoch:02d}-{val_accuracy:.2f}.h5' - 如果这不起作用,那么我认为这不是问题文件路径.. 如果笔记本是公开的,你有笔记本的链接吗?
    猜你喜欢
    • 1970-01-01
    • 1970-01-01
    • 1970-01-01
    • 1970-01-01
    • 2017-01-06
    • 1970-01-01
    • 1970-01-01
    • 2020-04-26
    • 2019-10-03
    相关资源
    最近更新 更多