Tensorflow.keras not training and stoping with error 'No algorithm worked!' in Colab

user_10830922 0
Viewed 1868

I created a CNN model and it trained properly for the first and completed all epochs but when I closed the runtime and came back after a day and tried to retrain. It gives this error.

NotFoundError:  No algorithm worked!
 [[node sequential_2/conv2d_4/Conv2D (defined at <ipython-input-7-5d7257468056>:45) ]] [Op:__inference_train_function_3464]

Note: I am training this in google colab Here's the error code.

model.compile(optimizer='adam' , loss='categorical_crossentropy', metrics=['accuracy'])
model.fit(train_dataset,batch_size=16,epochs=40,verbrose = 1,
         validation_data=val_data,validation_steps = 16 
         ,shuffle=True, steps_per_epoch=80,
         use_multiprocessing=False,callbacks=[callbacks])
model.save('/content/drive/My Drive/eyes2.h5')
model.summary()

Link to notebook https://colab.research.google.com/drive/1Qf05-Axd1gjuZhBDv7LwNtO_znIg4iKf

google-colaboratory keras tensorflow machine-learning python
2 Answers

I had this exact error message with Tensorflow when running a known-working model. As suggested by Bharath, it did appear to be related to version incompatibility.

However, I did not downgrade to fix it as this seems counter-intuitive and a step backwards. Here's an alternative solution that worked for me:

$ python3 -m pip install --upgrade pip
$ python3 -m pip install --upgrade pandas numpy matplotlib # probably not necessary
$ python3 -m pip uninstall tensorflow-gpu # if you have the old style packages installed
$ python3 -m pip install --upgrade tf-nightly

The Tensorflow installation instructions and GPU installation instructions were useful when solving this.

user_61515490

Try downgrading tensorflow to 1.x. Recently colab has updated tensorflow from 1.x to 2.x as its default which might be leading to this error. .

user_48006520
Related