ERR: DeadlineExceeded - Job was active longer than specified deadline

@janjongboom I tried the procedures on the tutorial for the Water Faucet Sound Classification Tutorial using my use case rsiquijor-project-1 (Sound classification of Ambient sound, motor vehicle, and an alert buzzer sound). At first I tried with 2 labels (Environmental Noise vs Motor Vehicle) and trying to add a 3rd one (Alert) sound. In process of doing so, I added the additional datasets and label them accordingly, but whenever I try to create an Impulse using MFCC, I get this error on the regeneration of the features, “ERR: DeadlineExceeded - Job was active longer than specified deadline”. Can you please point me to direction on what I’m doing wrong.

Thanks in advance!

@rsiquijor Can you up your window increase setting a bit (on the create impulse screen)? This reduces the time the process will take (limited to 10 minutes right now). Should not have a big effect on the performance of your ML model.

@janjongboom yes, I figured that was my problem. I completed it now. Thanks!

1 Like

@janjongboom Hi, I am having same problem. I’m trying to use transfer learning but I keep getting this error (maybe due to huge dataset). I can’t find where to change window size… Can you help me with this?

Hi @bhmin93,

The window size is used for time-serie data, it doesn’t apply to image classification.

Are you having this issue while training the neural network? Regarding the size of dataset, 50-100 samples for each class should be enough with Transfer Learning.


I’d also be interested to learn how many images are in your dataset, and what size they are—perhaps we need to up our time limits.

Until then, you can try uploading a randomly sampled subset of your dataset to get a sense for whether the problem is feasible to solve. This will also make training faster, so you can experiment more easily. Once we’ve figured out the timeout issue you can train with the full dataset.

@bhmin93, I’ve upped the compute time limit to 60 minutes for this project, let me know if that works!

@dansitu A bit over 10K images, scaled down to 96x96.