change dataset to cifar100

[axiniu]

@hyhieu Hi,Thank you for all you work. Now,I want to take place of the cifar10 dataset with cifar100 to implement this code to find and tune CNN cells and architectures. But I can't find the interface between dataset and network. Would you please give some advice? Thank you very much.

[MattVil]

look at src/cifar10/data_utils.py and implement your own read_data() function And you have to change some hard coded dimensions in:

micro_child.py: - build_valid_rl() - _model()

models.py: - _pre_process()

mainly

[axiniu]

@MattVil Thank you for you help,I have successfully implement this code to search for good cells with cifar100 dataset. But I still can't run it to search entire convolutional networks.Would you give me more advice? Thank you very much.

[leelaylay]

Change those lines in enas\src\cifar100\general_child.py or models\enas\src\cifar100\micro_child.py

- w = create_weight("w", [inp_c, 10])
+ w = create_weight("w", [inp_c, 100])

If the input shape changes, it is necessary to modify the _pre_process() in models.py.

[mumair5393]

Hi, I wanted to know while searching for micro cell for cifar100, does the validation and test accuracy improves over time. I am getting extremely low accuracy even if I train for longer period.