gcloud machine learning error compatibility keras incompatibility

Question

I'm working with machine learning on gcloud using SDK on my local terminal.

I'm running the following command:

gcloud ai-platform jobs submit training segmentation_maskrcnn_test_16 \
    --runtime-version 2.1 \
    --python-version 3.7 \
    --job-dir=gs://image-segmentation-meat/training_process \
    --package-path ./object_detection \
    --module-name object_detection.model_main_tf2 \
    --region us-central1 \
    --scale-tier CUSTOM \
    --master-machine-type n1-highcpu-32 \
    --master-accelerator count=8,type=nvidia-tesla-k80 \
    -- \
    --model_dir=gs://image-segmentation-meat/training_process \
    --pipeline_config_path=gs://image-segmentation-meat/mask_rcnn_inception_resnet_v2_1024x1024_coco17_gpu-8.config

But I got several errors like this:

I used this command to fix the problem direct on my terminal:

conda install -c conda-forge keras-preprocessing==1.1.0

I'm using miniconda on ubuntu 20.02, but didn't work. How and where I fix this?

Answer 1

In the end it was a simple error, I just needed to change the comand line on my terminal.

    gcloud ai-platform jobs submit training segmentation_maskrcnn_test_16 \
        --runtime-version 2.5 \
        --python-version 3.7 \
        --job-dir=gs://image-segmentation-meat/training_process \
        --package-path ./object_detection \
        --module-name object_detection.model_main_tf2 \
        --region us-central1 \
        --scale-tier CUSTOM \
        --master-machine-type n1-highcpu-32 \
        --master-accelerator count=8,type=nvidia-tesla-k80 \
        -- \
        --model_dir=gs://image-segmentation-meat/training_process \
        --pipeline_config_path=gs://image-segmentation-meat/mask_rcnn_inception_resnet_v2_1024x1024_coco17_gpu-8.config

So I changed --runtime-version 2.1 with --runtime-version 2.5.