Autokeras: How to use auto-keras for a binary classification task

[alokgogate]

Tried using this for Mnist dataset and for other image classification task, however wanted to use this for other regular binary classification tasks

Any documentation for the same?

I did have a look at the supervised main class and tried to use SearchSupervised

But kept on getting the following error:

TypeError: Can't instantiate abstract class SearchSupervised with abstract methods evaluate, final_fit, fit, predict

Any sample that I can refer to for a Binary Classification task (not image classification)?

Thanks!

[xujiameng]

I have the same problem and hope this issues can be answered.

[xujiameng]

I've considered converting two-dimensional arrays into four-dimensional data, which is then processed like image classification, but I don't think there will be any good results.

[christian-steinmeyer]

Does this help?

In essence, there, you'll find the following minimal example, assuming you have 'X_train, X_test, y_train, y_test' in the correct format.

from autokeras import MlpModule
from autokeras.backend.torch.loss_function import classification_loss
from autokeras.nn.metric import Accuracy
from autokeras.backend.torch import DataTransformerMlp

# ... define train and test data

mlpModule = MlpModule(loss=classification_loss, metric=Accuracy, searcher_args={}, verbose=True)
data_transformer = DataTransformerMlp(x_train)
train_data = data_transformer.transform_train(x_train, y_train)
test_data = data_transformer.transform_test(x_test, y_test)
fit_args = {
    "n_output_node": 2,
    "input_shape": x_train.shape,
    "train_data": train_data,
    "test_data": test_data
}
mlpModule.fit(n_output_node=fit_args.get("n_output_node"),
              input_shape=fit_args.get("input_shape"),
              train_data=fit_args.get("train_data"),
              test_data=fit_args.get("test_data"),
              time_limit=1 * 60 * 60)

Disclaimer: I've not been able to run this myself, as I have installation problems, but this is what I found and assume to work.

[alokgogate]

@christian-steinmeyer getting this error

Process ForkProcess-2:
Traceback (most recent call last):
File "/root/anaconda3/lib/python3.6/multiprocessing/process.py", line 258, in _bootstrap
self.run()
File "/root/anaconda3/lib/python3.6/multiprocessing/process.py", line 93, in run
self._target(self._args, *self._kwargs)
File "/root/anaconda3/lib/python3.6/site-packages/autokeras/search.py", line 351, in train
path=path,
File "/root/anaconda3/lib/python3.6/site-packages/autokeras/search.py", line 344, in train
File "/root/anaconda3/lib/python3.6/site-packages/autokeras/nn/model_trainer.py", line 137, in train_model
File "/root/anaconda3/lib/python3.6/site-packages/autokeras/nn/model_trainer.py", line 174, in _train
File "/root/anaconda3/lib/python3.6/site-packages/autokeras/backend/torch/loss_function.py", line 5, in classification_loss
labels = target.argmax(1)
File "/root/anaconda3/lib/python3.6/site-packages/torch/tensor.py", line 240, in argmax
return torch.argmax(self, dim, keepdim)
File "/root/anaconda3/lib/python3.6/site-packages/torch/functional.py", line 544, in argmax
retained or not. Ignored if dim=None.
RuntimeError: Dimension out of range (expected to be in range of [-1, 0], but got 1)

[LukasKapp-Schwoerer]

@alokgogate I had the same error because I didn't pass the labels one-hot-encoded.

[huaiyizhao]

I'm not sure if it is the appropriate way to do classification since MlpModel's output is a float rather than {0,1}, the "classification" loss function may not apply here.

[christian-steinmeyer]

@alokgogate I had the same error because I didn't pass the labels one-hot-encoded.

@LukasKapp-Schwoerer's suggestion should solve the problem of @alokgogate.

@huaiyizhao Could you elaborate on your reply? Perhaps have a look at #678, there, I've further explained another problem I have. In essence, the loss is undefined such that autokeras does not converge.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.