Where communities thrive

  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
Mehmet Sonmez
Hi everyone, I am trying to use Autokeras to build a model that takes an input sequence of shape (150000,1) and outputs 1 value. Essentially a regression task. What other modules are there besides ImageClassifier? I was looking at the documentation but couldn't find any other examples and couldn't find any other modules directly in the codebase
Juan C Rodriguez
Hi Mehmet,
the different modules will depend on the Auto-Keras version you have installed. If you are using the pip version (0.3.7) then you can try:
ImageClassifier , ImageRegressor , TextRegressor , TextClassifier , TabularRegressor , TabularClassifier
Mehmet Sonmez
Thank you Juan : ) What does Tabular regressor do? I'm familiar with image and text but am unsure of what tabular refers to here
Juan C Rodriguez
Well, I'm not quite sure what the tabular modules are for since I haven't used them, and I don't want to lie.
However, don't get too carried away with them, as the developers of Auto-Keras announced that they are going to remove them (they already did it in the development branch), and they are going to include them as a parallel project.

I am trying to run the mlp module script given in the autokeras net modules examples.

I noticed that the search process stops after the fit function hits the specified time limit. Is there a way to instead specify the number of different models you'd want the searcher to look through.

hello there, I'm not able to export the auokeras model into keras. I get this error. Traceback (most recent call last):
File "autoML.py", line 64, in <module>
File "/usr/local/lib/python3.6/dist-packages/autokeras-0.3.7-py3.6.egg/autokeras/supervised.py", line 252, in export_keras_model
AttributeError: 'Graph' object has no attribute 'produce_keras_model'
Can someone please help? Thanks
Also, not able to execute load searcher
I have latest version of autokeras
Jezabel R. Garcia
Hi! Have someone used the TabularRegressor tool? Also, I can not find the tutorials in the webpage
Hi experts,
Is it possible to use AutoKeras with "flow_from_directory" method in Keras and how can I implement that? I tried to do K-fold CV and Hyperas, but all of them need CSV labeling file.
@shivaji27 I have got the same problem!
Nathan Liu
hello everybady ,I get a problem when wanna see the autokeras model,
if name == 'main':
AttributeError: 'NoneTyle' object has no attribute 'path'
Can somebady help me? thank you
Nathan Liu
I can not use the file named visualize.py
I know autokeras can predict classes using results = clf.predict(x_test), but is there another function similar to predict_proba that can also provide the confidence of those predictions?
Juan Park
Hello everyone, like @HYDesmondLiu, I am also looking at if there is an implementation for flow_from_directory or flow_from_dataframe.
Juan Park

Also, does anyone happen to have a code example of training a multilabel model?
I am feeding the ImageClassifier, trainX (of shape: (n_images, 199, 199, 3)) and trainY (of shape: (n_images, n_classes)).

    model = ak.ImageClassifier(verbose=True)
    model.fit(trainX, trainY, time_limit=seconds)

I am getting the error:

ValueError: Found input variables with inconsistent numbers of samples: [16933, 474124]

Where 474124 is n_classes*n_images. To me, it looks ImageClassifier assumes the labels to be in multiclass format (ie. class=[0, 1, 2, 3, 4....]).

Sinan Sivri
Hello everyone best practice is using autokeras with docker? i'm unable to install via pip
could you help?
Hi - I was wondering if anyone had any advice on how to use autokeras with multiple inputs?
for context, I am working on a vector classification problem (i.e. my output is a vector of non-softmax binary classifiers)
and I have a text input and an audio input
I'd like to keep them separate as the text input is non-embedded tokens (and would need to be embedded first somehow), so I don't think I can just concatenate them
concatenate text and audio input, I mean
@jhfjhfj1 is this channel monitored?
Huaiyi Zhao
Hi all.
Hi everyone. I am wondering how to use the funcion ''visualize’’ to see what a model looks like since i do not know where exactly can i get this function, more specificly, what part do i need to import?

Hi All,
Please you help me.
I try to use Supervised Class. it occur error TypeError as picture.
How to use Supervised Class.


Vijay Daita
I was wondering if you could use AutoKeras as a replacement for AutoML
How would it compare to something like fast.ai?
I can't find the example of cifar10 (keras-team/autokeras@85d972a), where is it?
anyone here?
How many epochs does it run on each model? And does Current Metric = train or test accuracy?
can someone help me integrate autokeras in my app?
Hi All,
Does anybody have any experience using AutoKeras with time series data?
I also would like to get some tips on which cuda / tensorflow / autokeras versions work best, trying to setup a docker env so I can share it with the people on my team
Fan Yang
Hi everybody,
I got this error"RuntimeError: Given groups=1, weight of size [2048, 6144, 1], expected input[128, 8192, 128] to have 6144 channels, but got 8192 channels instead
how to solve this?
Elad Rapaport
can autokeras be used for "image" data with more than 3 channels? I get "RuntimeError: running_mean should contain 3 elements not 22" when trying to run such a dataset but I was hoping that there is a tweak that can allow this?
Sravya Tirukkovalur
Hi all, I am looking into the code base of autokeras and reading through the paper, seems like APIs have changed quite a bit recently( may be after 0.4.0 release). For example: no more search.py, bayesian.py . Curious, when is the next release planned? Asking to understand if it might be better to wait to dig through the code for the next release.
Tomasz Latkowski
Hi all, i have a general question:
Does anyone try to train auto-keras on large datasets i.e. ImageNet?
I'm wondering how the results look like in comparison to RL-based NAS approaches
hello?can autokeras be used for "image" data with more than 3 channels?