keras-team/autokeras

AutoML library for deep learning

Official Website: autokeras.com

AutoKeras: An AutoML system based on Keras. It is developed by DATA Lab at Texas A&M University. The goal of AutoKeras is to make machine learning accessible to everyone.

Learning resources

• A short example.

import autokeras as ak

clf = ak.ImageClassifier()
clf.fit(x_train, y_train)
results = clf.predict(x_test)

• Official website tutorials.
• The book of Automated Machine Learning in Action.
• The LiveProjects of Image Classification with AutoKeras.

   

Installation

To install the package, please use the pip installation as follows:

pip3 install autokeras

Please follow the installation guide for more details.

Note: Currently, AutoKeras is only compatible with Python >= 3.7 and TensorFlow >= 2.8.0.

Community

Ask your questions on our GitHub Discussions.

Contributing Code

Here is how we manage our project.

We pick the critical issues to work on from GitHub issues. They will be added to this Project. Some of the issues will then be added to the milestones, which are used to plan for the releases.

Refer to our Contributing Guide to learn the best practices.

Thank all the contributors!

Cite this work

Haifeng Jin, François Chollet, Qingquan Song, and Xia Hu. "AutoKeras: An AutoML Library for Deep Learning." the Journal of machine Learning research 6 (2023): 1-6. (Download)

Biblatex entry:

@article{JMLR:v24:20-1355,
  author  = {Haifeng Jin and François Chollet and Qingquan Song and Xia Hu},
  title   = {AutoKeras: An AutoML Library for Deep Learning},
  journal = {Journal of Machine Learning Research},
  year    = {2023},
  volume  = {24},
  number  = {6},
  pages   = {1--6},
  url     = {http://jmlr.org/papers/v24/20-1355.html}
}

Acknowledgements

The authors gratefully acknowledge the D3M program of the Defense Advanced Research Projects Agency (DARPA) administered through AFRL contract FA8750-17-2-0116; the Texas A&M College of Engineering, and Texas A&M University.

autokeras's Issues

GPU requirement?

Thank you for sharing your project sir.

I am about to use this source for my journal paper.
Original papers said that the authers used several hundreds of graphic card to run AutoML.

How about this source?
Ex: in my case, i am using only one Titan X graphic card with 5000 images (224x224x3).
Actually i already got memory error. What is minimum requirement of graphic card to run this source sir?

Website SSL certification is wrong

Cool project!

But while I tried to accesso https://autokeras.com/ I'll receive an SSL error from browser, the certification is issued for github only.

You may change the configuration from website's repo at:

and fill autokeras.com into that field.

ImportError: cannot import name 'abs'

I am trying the example mnist.py but repeatedly getting this error. I have uninstalled tensorflow and reinstalled it's latest version but the problem still persists. How can I rectify this? P.S. all solutions I found online were for Windows, I am using Ubuntu 18.04

from: can't read /var/mail/keras.datasets

Describe the bug:
Running an simple MNIST data classification.
I was able to run for the first time. Next day, I tried to run the same program, I got this error message

(asr) maybe@maybe1:/data/jugs/asr/languageclf$ ./trainalutokeras.py 
from: can't read /var/mail/keras.datasets
from: can't read /var/mail/autokeras.classifier
./trainalutokeras.py: line 5: syntax error near unexpected token `,'
./trainalutokeras.py: line 5: `  (x_train, y_train),(x_test, y_test) = mnist.load_data()'

Setup Details:
Any details about your local setup that might be helpful in troubleshooting. Include the details about the versions of:

• OS type and version: Ubuntu 16.04
• Python: 3.6
• scikit-learn: 0.19
• numpy: 1.14.x
• keras: 2.1.6
• scipy: 1.0.1
• tensorflow: 1.6

Autokeras installation fails

pip install autokeras Collecting autokeras Using cached https://files.pythonhosted.org/packages/cf/3f/ef52a7654445180ac174b6c159e44261e4fa4161bb28973229845deeedee/autokeras-0.2.0.tar.gz Collecting torch==0.4.0 (from autokeras) Could not find a version that satisfies the requirement torch==0.4.0 (from autokeras) (from versions: 0.1.2, 0.1.2.post1) No matching distribution found for torch==0.4.0 (from autokeras)

Windows 10
Python 3.6.2

ImageClassifier3D

Feature Description

Your library can do 3D convolutional auto keras.

Reason

Input images are 3D images, such as 3D CT scan images.

Solution

If you add this feature, it will benefit many people.

Alternative Solutions

Benchmark Datasets

3D-MINST

Additional Context

Failed to run examples/mnist.py

Running examples/mnist.py
and got the following error:

...
Father ID: 2
[('to_wider_model', 1, 64), ('to_wider_model', 1, 128), ('to_wider_model', 1, 256), ('to_wider_model', 1, 512), ('to_conv_deeper_model', 6, 3), ('to_add_skip_model', 6, 19), ('to_conv_deeper_model', 1, 5), ('to_conv_deeper_model', 11, 5), ('to_wider_model', 19, 64), ('to_wider_model', 11, 64), ('to_conv_deeper_model', 25, 3), ('to_add_skip_model', 19, 11), ('to_wider_model', 1, 1024), ('to_conv_deeper_model', 33, 3)]
Traceback (most recent call last):
File "mnist.py", line 12, in
clf.fit(x_train, y_train, time_limit=12 * 60 * 60)
File "/home/fayw/anaconda3/lib/python3.6/site-packages/autokeras/classifier.py", line 219, in fit
run_searcher_once(train_data, test_data, self.path)
File "/home/fayw/anaconda3/lib/python3.6/site-packages/autokeras/classifier.py", line 41, in run_searcher_once
searcher.search(train_data, test_data)
File "/home/fayw/anaconda3/lib/python3.6/site-packages/autokeras/search.py", line 168, in search
new_graph, new_father_id = self.maximize_acq()
File "/home/fayw/anaconda3/lib/python3.6/site-packages/autokeras/search.py", line 229, in maximize_acq
getattr(nm_graph, args[0])(*list(args[1:]))
File "/home/fayw/anaconda3/lib/python3.6/site-packages/autokeras/graph.py", line 311, in to_conv_deeper_model
new_layers = deeper_conv_block(target, kernel_size, self.weighted)
File "/home/fayw/anaconda3/lib/python3.6/site-packages/autokeras/layer_transformer.py", line 11, in deeper_conv_block
n_filters = conv_layer.filters
AttributeError: 'StubDropout' object has no attribute 'filters'

Please help!

Export the autokeras model

Feature Description

Export the Autokeras model of its own type instead of exporting the Keras model.

Reason

More flexible than Keras model.

Solution

Create a base class which can only be loaded and do the "predict" and "evaluation" function.
A class named "PortableImageClassifer" extends the base class.
It has a member of an instance of "Graph".
Refer to other parts of the project to see how to do "predict" and "evaluation" with an instance of Graph.

Alternative Solutions

NA

Additional Context

NA

PermissionError: [Errno 13] Permission denied: '/tmp/autokeras/searcher'

I've encountered with this error
File "/home/me/venv/lib64/python3.6/site-packages/autokeras/classifier.py", line 258, in save_searcher
pickle.dump(searcher, open(os.path.join(self.path, 'searcher'), 'wb'))
PermissionError: [Errno 13] Permission denied: '/tmp/autokeras/searcher'

I'm running this on linux server on virtual environment. Could you please give me the suggestion?

overflow encountered in exp on bayesian.py line 89

Bug Description

When running image_classifier. If we use a big MAX_MODEL_NUM (1000). https://github.com/jhfjhfj1/autokeras/blob/master/autokeras/bayesian.py#L89 would throw an overflow warning. Which causes a FileNotFound Error to be raised.

Here is the stack trace:

  File "/home/tsunglin/autokeras/autokeras/image_classifier.py", line 229, in fit
    run_searcher_once(train_data, test_data, self.path, int(time_remain))
  File "/home/tsunglin/autokeras/autokeras/image_classifier.py", line 42, in run_searcher_once
    searcher.search(train_data, test_data, timeout)
  File "/home/tsunglin/autokeras/autokeras/search.py", line 189, in search
    new_graph, new_father_id = self.maximize_acq(timeout)
  File "/home/tsunglin/autokeras/autokeras/search.py", line 268, in maximize_acq
    nm_graph = self.load_model_by_id(father_id)
  File "/home/tsunglin/autokeras/autokeras/search.py", line 109, in load_model_by_id
    return pickle_from_file(os.path.join(self.path, str(model_id) + '.h5'))
  File "/home/tsunglin/autokeras/autokeras/utils.py", line 190, in pickle_from_file
    return pickle.load(open(path, 'rb'))
FileNotFoundError: [Errno 2] No such file or directory: 'tests/resources/temp/None.h5'

Reproducing Steps

Steps to reproduce the behavior:

• Step 1: open test_image_classifier
• Step 2: change Constant.MAX_MODEL_NUM to 1000 in test_save_continue function.
• Step 3: run the unit test and you'll see the exception.

Expected Behavior

Without error

Setup Details

Include the details about the versions of:

• OS type and version: Ubuntu 16.04
• Python: 3.6
• autokeras: commit f786818
• scikit-learn: 0.19.1
• numpy:1.14.5
• keras: 2.2.2
• scipy:
• tensorflow: 1.8.0
• pytorch: 0.4.0

Additional context

Parallelism

I have your /examples/mnist.py file working on an AWS GPU server. I needed to delete the "time_limit=12 * 60 * 60" part though because that was giving an error. My question is, does it use multiple cores by default? I read the Parallelism section of your paper, but am unclear if I have to do anything to make it parallel. What if for example I want to run it on a CPU-only server with 128 cores, would it automatically use all of them? Also, when you say it runs multi-GPU, do you mean within the same machine, or can it also run over a cluster of multiple GPU servers?

Also, in Table 1 of your paper, where you compare Auto-Keras with similar programs, are those accuracy results on the test set or validation set? In your github repo it seems to be only split into training and test sets. Should that also include a validation set?

may you share code for your papers

it would be great to share a code for
Mining Mobile Internet Packets for Malware Detection

Multi-layer Anomaly Detection for Internet Traffic Based on Data Mining

thanks

Possible to use in regression problem?

I have a dataset of houses and sale prices that I'm playing around with, is autokeras capable of using for regression problems, or will it be in the future?

TextClassifier

I see ImageClassifier in the doc. Is there something like TextClassifier to process text?

GPU usage

Your /examples/mnist.py does not load GPUtil like your older experiments files do (like
https://github.com/jhfjhfj1/autokeras/blob/master/experiments/mnist.py ). Does it no longer need GPUtil?
It was not on your requirements.txt list.

This is what I am talking about:

import os
import GPUtil

def select_gpu():
try:
# Get the first available GPU
DEVICE_ID_LIST = GPUtil.getFirstAvailable()
DEVICE_ID = DEVICE_ID_LIST[0] # grab first element from list

    # Set CUDA_VISIBLE_DEVICES to mask out all other GPUs than the first available device id
    os.environ["CUDA_VISIBLE_DEVICES"] = str(DEVICE_ID)
except FileNotFoundError:

print("GPU not found")

Do you support rnn structure?

I am working on rnn predicting by keras. And I want to know if this autokeras can support it?

clf.fit(x_train, y_train, time_limit=1) running for a long time on mnist

Not sure if the argument time_limit is used here. This function is running for a very long time.

clf.fit(x_train, y_train, time_limit=1)

Any possible reason?

Mac:

Python 3ish

load_image_dataset

autokeras.classifier.load_image_dataset failed silently

x_train, y_train = load_image_dataset(csv_file_path="***_train.csv", images_path=".")
print(x_train.shape)
print(y_train.shape)

***_train.csv contains 100k lines of image path and labels

I try to do load_image_dataset, and wish to get x_train.shape. However, it doesn't print any value.
so is the load_image_dataset not able to handle 100k data size?

Parameter Usage

How do I use the parameters? If I use
clf = ImageClassifier(verbose=True, searcher_args={'trainer_args': {'max_iter_num': 10, 'augment': False}, 'default_model_len': 3})

I get this error: TypeError: init() got an unexpected keyword argument 'searcher_args'

Out of memory error with NVIDIA K80 GPU

Trying to create an image classifier with ~1000 training samples and 7 classes but it throws a runtime error. Is there a way of reducing batch size or something else that can be done to circumvent this?

Following is the error.

RuntimeError: cuda runtime error (2) : out of memory at /pytorch/aten/src/THC/generic/THCStorage.cu:58/usr/lib/python3.5/multiprocessing/semaphore_tracker.py:129: UserWarning: semaphore_tracker: There appear to be 2 leaked semaphores to clean up at shutdown len(cache))

Possible to use for 3D convolutioanl neural network?

You only support the image classification problems. Am I right? If my problem is similar to the image classification, but my input images are 3D instead of 2D. Can I still use your AutoKeras to figure out the better architectures and hyperparameters for 3D convolutional neural networks? Thanks for sharing your library.

ObjectDetection

Task Module: Object Detection

Task Description

• Input: one image or more, each of them contains one or more objects need to be detected.
• Output: for each input image, this module output one or more bounding boxes of objects to be detected, and the class name they belong to.

Evaluation Metrics: mAP (mean Average Precision)

1. For each detected object, the bounding box IoU (intersection over union) is calculated, which is between 0~1
2. The detection is accepted iff: 1) classification is correct, 2) the IoU is great or equal to the IoU threshold
3. If there exists multiple IoU thresholds, then the precision is averaged over all thresholds
4. For each class we calculate its average precision (AP), and then we average all classes to get the mAP

Benchmark Datasets

1. PASCAL Visual Object Classification (PASCAL VOC) dataset
2. Common Objects in COntext (COCO) dataset

Reason

Object detection is one of the most important research areas and applications in Computer Vision, such as face detection, vehicle detection, people counting. This module will provide easy access to people or group who needs to perform object detection in their work with limited Computer Vision or Machine Learning background.

Solution

Some add-ons on current AutoKeras to be considered:

1. Add RoI layer into search space and IoU loss function.
2. Consider using some popular architectures as initializations (e.g. SSD, YoLo).
3. Consider adding complexity constraint (e.g. depth, width) in searching metrics. Currently the accuracy is the only metrics for the model selection.

RuntimeError: Given groups=1, weight[64, 4, 3, 3], so expected input to have 4 channels, but got 1 channels instead

Hi, my input tensor was in shape [batch, 20, 20, 1] and got this error:

Traceback (most recent call last):
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/multiprocessing/pool.py", line 119, in worker
result = (True, func(*args, **kwds))
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/multiprocessing/pool.py", line 44, in mapstar
return list(map(*args))
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/site-packages/autokeras/search.py", line 327, in train
verbose).train_model(**trainer_args)
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/site-packages/autokeras/utils.py", line 122, in train_model
self._train(train_loader, epoch)
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/site-packages/autokeras/utils.py", line 143, in _train
outputs = self.model(inputs)
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/site-packages/autokeras/graph.py", line 603, in forward
temp_tensor = torch_layer(edge_input_tensor)
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/site-packages/torch/nn/modules/module.py", line 491, in call
result = self.forward(*input, **kwargs)
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 301, in forward
self.padding, self.dilation, self.groups)
RuntimeError: Given groups=1, weight[64, 4, 3, 3], so expected input[128, 1, 20, 20] to have 4 channels, but got 1 channels instead
"""

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File "autokeras_alter.py", line 35, in
clf.fit(X_train, y_train, time_limit=246060)
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/site-packages/autokeras/classifier.py", line 225, in fit
run_searcher_once(train_data, test_data, self.path)
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/site-packages/autokeras/classifier.py", line 40, in run_searcher_once
searcher.search(train_data, test_data)
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/site-packages/autokeras/search.py", line 190, in search
accuracy, loss, graph = train_results.get()[0]
File "/public/users/liangqiaoxing/.conda/envs/tensorflow_py3.6/lib/python3.6/multiprocessing/pool.py", line 644, in get
raise self._value
RuntimeError: Given groups=1, weight[64, 4, 3, 3], so expected input[128, 1, 20, 20] to have 4 channels, but got 1 channels instead

---

The error message seemed to come from Torch. I tried running the same ImageClassifier code on mnist data with shape [batch, 28, 28, 1] and the code worked fine.

Here is my training set. Does anyone know why or meet the same error? Thanks!
demo.zip

Three undefined names in pytorch_cifar10.py

flake8 testing of https://github.com/jhfjhfj1/autokeras on Python 3.7.0

$ flake8 . --count --select=E901,E999,F821,F822,F823 --show-source --statistics

./experiments/pytorch_cifar10.py:51:7: F821 undefined name 'ResNet18'
net = ResNet18()
      ^
./experiments/pytorch_cifar10.py:98:9: F821 undefined name 'progress_bar'
        progress_bar(batch_idx, len(trainloader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
        ^
./experiments/pytorch_cifar10.py:118:13: F821 undefined name 'progress_bar'
            progress_bar(batch_idx, len(testloader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
            ^
3     F821 undefined name 'ResNet18'
3

Things to add to CONTRIBUTING.md

@tl-yang Please add 2 items to the pull request guide in the file.

1. submit the pull request from the first day you started your development. mark with WIP.
2. When the status change from WIP to MRG, assign the reviewer to jhfjhfj1.

An issue on multiprocessing with CUDA

The example code on MNIST showed an issue on multiprocessing with CUDA.
I guess CUDA doesn't work well with the multiprocessing https://pytorch.org/docs/stable/notes/windows.html
I use Windows10, Python 3.6.5, CUDA 9.0, and tensorflow-gpu==1.8.0. I followed requirements.txt to install others.
Using commands "torch.version.cuda" and "torch.cuda.is_available()", I also checked CUDA 9.0 works well on my environment.

...........................................
Epoch 1: loss 1.8169708251953125, metric_value 0.987
...........................................
Epoch 2: loss 1.694014310836792, metric_value 0.9866
...........................................
Epoch 3: loss 1.5495505332946777, metric_value 0.9884
...........................................
Epoch 4: loss 1.7369670867919922, metric_value 0.9882
...........................................
Epoch 5: loss 2.2431726455688477, metric_value 0.9842
...........................................
Epoch 6: loss 1.821488618850708, metric_value 0.9864
...........................................
Epoch 7: loss 1.3172129392623901, metric_value 0.9908
...........................................
Epoch 8: loss 1.3580961227416992, metric_value 0.9906
...........................................
Epoch 9: loss 1.2147481441497803, metric_value 0.992
...........................................
Epoch 10: loss 1.3565152883529663, metric_value 0.9916
...........................................
Epoch 11: loss 1.201681137084961, metric_value 0.9916
...........................................
Epoch 12: loss 1.9307321310043335, metric_value 0.987
...........................................
Epoch 13: loss 1.2660953998565674, metric_value 0.993
...........................................
Epoch 14: loss 1.2389497756958008, metric_value 0.9924
...........................................
Epoch 15: loss 1.3451954126358032, metric_value 0.992
...........................................
Epoch 16: loss 2.506535291671753, metric_value 0.9844
...........................................
Epoch 17: loss 2.018087387084961, metric_value 0.989
...........................................
Epoch 18: loss 1.1533660888671875, metric_value 0.9934
...........................................
Epoch 19: loss 1.493323802947998, metric_value 0.9908
No loss decrease after 5 epochs
THCudaCheck FAIL file=c:\programdata\miniconda3\conda-bld\pytorch_1533090623466\work\torch\csrc\generic\StorageSharing.cpp line=231 error=63 : OS call failed or operation not supported on this OS
Traceback (most recent call last):
File "ak_mnist.py", line 10, in
clf.fit(x_train, y_train, time_limit=12 * 60 * 60)
File "C:\Users\kmbmjn\Anaconda3\lib\site-packages\autokeras\image_supervised.py", line 238, in fit
run_searcher_once(train_data, test_data, self.path, int(time_remain))
File "C:\Users\kmbmjn\Anaconda3\lib\site-packages\autokeras\image_supervised.py", line 41, in run_searcher_once
searcher.search(train_data, test_data, timeout)
File "C:\Users\kmbmjn\Anaconda3\lib\site-packages\autokeras\search.py", line 189, in search
metric_value, loss, graph = train_results.get(timeout=remaining_time)[0]
File "C:\Users\kmbmjn\Anaconda3\lib\multiprocessing\pool.py", line 644, in get
raise self._value
multiprocessing.pool.MaybeEncodingError: Error sending result: '[(0.9899199999999999, tensor(1.7033, device='cuda:0'), <autokeras.graph.Graph object at 0x00000174DB1BC7B8>)]'. Reason: 'RuntimeError('cuda runtime error (63) : OS call failed or operation not supported on this OS at c:\programdata\miniconda3\conda-bld\pytorch_1533090623466\work\torch\csrc\generic\StorageSharing.cpp:231',)'

Performance

Hi --

Do you have any information about the expected performance of this project on standard benchmarks, eg MNIST or CIFAR10?

Thanks

ImportError: cannot import name 'Classifier'

Bug Description

from autokeras.image_classifier import ImageClassifier

This gives the following error:

ImportError Traceback (most recent call last)
in ()
1 from keras.datasets import mnist
----> 2 from autokeras.image_classifier import ImageClassifier

/usr/local/lib/python3.6/dist-packages/autokeras/image_classifier.py in ()
18 from autokeras.search import BayesianSearcher, train
19 from autokeras.utils import ensure_dir, has_file, pickle_from_file, pickle_to_file
---> 20 from autokeras.classifier import Classifier
21
22

ImportError: cannot import name 'Classifier'

Reproducing Steps

Steps to reproduce the behavior:
Use this https://colab.research.google.com/drive/1it4U3H8t6fMhLBt_NFmB6vRRN2bVpxFv

Expected Behavior

Separating classifier.py and image_classifier.py has created this problem.

Setup Details

Include the details about the versions of:

• OS type and version: Linux-4.14.33+-x86_64-with-Ubuntu-17.10-artful
• Python: 3.6.3
• scikit-learn: 0.19.1
• numpy: 1.14.5
• keras: 2.2.2
• scipy: 0.19.1
• tensorflow: 1.10.0
• pytorch: 0.4.0

TypeError: 'float' object cannot be interpreted as an integer

Traceback (most recent call last):
File "C:/Users/lichunhong/Desktop/python_code/autokeras_try/1.py", line 10, in
clf.fit(x_train, y_train, time_limit=12 * 60 * 60)
File "C:\ProgramData\Anaconda3\lib\site-packages\autokeras\classifier.py", line 225, in fit
run_searcher_once(train_data, test_data, self.path)
File "C:\ProgramData\Anaconda3\lib\site-packages\autokeras\classifier.py", line 40, in run_searcher_once
searcher.search(train_data, test_data)
File "C:\ProgramData\Anaconda3\lib\site-packages\autokeras\search.py", line 190, in search
accuracy, loss, graph = train_results.get()[0]
File "C:\ProgramData\Anaconda3\lib\multiprocessing\pool.py", line 644, in get
raise self._value

TypeError: 'float' object cannot be interpreted as an integer

cifar10 example fails

C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\site-packages\h5py\__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from floattonp.floatingis deprecated. In future, it will be treated asnp.float64 == np.dtype(float).type`.
from ._conv import register_converters as _register_converters
Using TensorFlow backend.
Initializing search.
Initialization finished.
Training model 0
multiprocessing.pool.RemoteTraceback:
"""
Traceback (most recent call last):
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\multiprocessing\pool.py", line 119, in worker
result = (True, func(*args, **kwds))
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\multiprocessing\pool.py", line 44, in mapstar
return list(map(*args))
File "C:\Users\Adam\Desktop\autokeras\examples\autokeras\search.py", line 327, in train
verbose).train_model(**trainer_args)
File "C:\Users\Adam\Desktop\autokeras\examples\autokeras\utils.py", line 122, in train_model
self._train(train_loader, epoch)
File "C:\Users\Adam\Desktop\autokeras\examples\autokeras\utils.py", line 143, in _train
outputs = self.model(inputs)
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "C:\Users\Adam\Desktop\autokeras\examples\autokeras\graph.py", line 603, in forward
temp_tensor = torch_layer(edge_input_tensor)
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\conv.py", line 301, in forward
self.padding, self.dilation, self.groups)
TypeError: 'float' object cannot be interpreted as an integer
"""

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File "cifar10.py", line 8, in
clf.fit(x_train, y_train, time_limit=12 * 60 * 60)
File "C:\Users\Adam\Desktop\autokeras\examples\autokeras\classifier.py", line 225, in fit
run_searcher_once(train_data, test_data, self.path)
File "C:\Users\Adam\Desktop\autokeras\examples\autokeras\classifier.py", line 40, in run_searcher_once
searcher.search(train_data, test_data)
File "C:\Users\Adam\Desktop\autokeras\examples\autokeras\search.py", line 190, in search
accuracy, loss, graph = train_results.get()[0]
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\multiprocessing\pool.py", line 644, in get
raise self._value
TypeError: 'float' object cannot be interpreted as an integer
`

Windows 10
Python 3.6.2

cifar10.py example fails.

layers_distance (f[len_a][len_b] is always equal to f[-1][-1]=0) ,right?

f[len_a][len_b] is always equal to f[-1][-1]=0

    len_a = len(list_a)
    len_b = len(list_b)
    f = np.zeros((len_a + 1, len_b + 1))
    f[-1][-1] = 0
    for i in range(-1, len_a):
        f[i][-1] = i + 1
    for j in range(-1, len_b):
        f[-1][j] = j + 1
    for i in range(len_a):
        for j in range(len_b):
            f[i][j] = min(f[i][j - 1] + 1, f[i - 1][j] + 1, f[i - 1][j - 1] + layer_distance(list_a[i], list_b[j]))
    print(f)
    return f[len_a][len_b]

A merge layer should be called on a list of inputs

Traceback (most recent call last):
File "example.py", line 50, in
test_my_layer()
File "example.py", line 39, in test_my_layer
b = WeightedAdd()(a)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 590, in call
self.build(input_shapes[0])
File "/usr/local/lib/python2.7/dist-packages/keras/layers/merge.py", line 68, in build
raise ValueError('A merge layer should be called '
ValueError: A merge layer should be called on a list of inputs.

Keras on Tensorflow 1.5

Don't we use oneHot in this source?

Hey mr

i got following error (the code is from classifier.py in this source):

if x_train.shape[0] != y_train.shape[0]:
raise ValueError('x_train and y_train should have the same number of instances.')

where x_train.shape is 972 but y_train.shape is 48600 which is 50 times than x_train.
because, i have 50 classes, so the oneHot gives me array of 50 numbers ([0, 0, 1, 0 .....0]) per image.

Link to autokeras.com

Please add a link to the docs at http://autokeras.com/ in the readme of this repo. Thanks.

TimeoutError when running mnist.py

Bug Description

When running mnist.py after some time, suddenly an error occurred :

Traceback (most recent call last):
File "/home/jdiez/project/AutoKeras/Test_01.py", line 11, in
clf.fit(x_train, y_train, time_limit=20 * 60)
File "/home/jdiez/lib/anaconda3/envs/autokeras/lib/python3.6/site-packages/autokeras/image_supervised.py", line 238, in fit
run_searcher_once(train_data, test_data, self.path, int(time_remain))
File "/home/jdiez/lib/anaconda3/envs/autokeras/lib/python3.6/site-packages/autokeras/image_supervised.py", line 41, in run_searcher_once
searcher.search(train_data, test_data, timeout)
File "/home/jdiez/lib/anaconda3/envs/autokeras/lib/python3.6/site-packages/autokeras/search.py", line 175, in search
timeout)
File "/home/jdiez/lib/anaconda3/envs/autokeras/lib/python3.6/site-packages/autokeras/bayesian.py", line 272, in optimize_acq
raise TimeoutError
TimeoutError

Process finished with exit code 1

Reproducing Steps

Steps to reproduce the behavior:

• Step 1: Follow the instruction steps for 0.2.4 install
• Step 2: Run mnist.py

Expected Behavior

Run without error

Setup Details

Include the details about the versions of:

• OS type and version: Ubuntu 16.04
• Python: 3.6
• autokeras: 0.2.4
• scikit-learn: 0.19.1
• numpy: 1.14.5
• keras: 2.2.2
• scipy: 1.1.0
• tensorflow: 1.8.0
• pytorch: 0.4.1 (py36_cuda9.2.148_cudnn7.1.4_1)

Additional context

Torch, Tensorflow and Keras are all working fine insisde this conda env.
The same error occurs when adapting the code for CIFAR10, Fashion-MNIST

mnist example fails

C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\site-packages\h5py\__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from floattonp.floatingis deprecated. In future, it will be treated asnp.float64 == np.dtype(float).type`.
from ._conv import register_converters as _register_converters
Using TensorFlow backend.
Initializing search.
Initialization finished.
Training model 0
multiprocessing.pool.RemoteTraceback:
"""
Traceback (most recent call last):
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\multiprocessing\pool.py", line 119, in worker
result = (True, func(*args, **kwds))
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\multiprocessing\pool.py", line 44, in mapstar
return list(map(*args))
File "C:\Users\Adam\Desktop\autokeras\autokeras\examples\autokeras\search.py", line 323, in train
verbose).train_model(**trainer_args)
File "C:\Users\Adam\Desktop\autokeras\autokeras\examples\autokeras\utils.py", line 121, in train_model
self._train(train_loader, epoch)
File "C:\Users\Adam\Desktop\autokeras\autokeras\examples\autokeras\utils.py", line 140, in _train
outputs = self.model(inputs)
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "C:\Users\Adam\Desktop\autokeras\autokeras\examples\autokeras\graph.py", line 577, in forward
temp_tensor = torch_layer(edge_input_tensor)
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\nn\modules\conv.py", line 301, in forward
self.padding, self.dilation, self.groups)
TypeError: 'float' object cannot be interpreted as an integer
"""

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
File "mnist.py", line 10, in
clf.fit(x_train, y_train, time_limit=12 * 60 * 60)
File "C:\Users\Adam\Desktop\autokeras\autokeras\examples\autokeras\classifier.py", line 224, in fit
run_searcher_once(train_data, test_data, self.path)
File "C:\Users\Adam\Desktop\autokeras\autokeras\examples\autokeras\classifier.py", line 40, in run_searcher_once
searcher.search(train_data, test_data)
File "C:\Users\Adam\Desktop\autokeras\autokeras\examples\autokeras\search.py", line 187, in search
accuracy, loss, graph = train_results.get()[0]
File "C:\Users\Adam\AppData\Local\Programs\Python\Python36\lib\multiprocessing\pool.py", line 644, in get
raise self._value
TypeError: 'float' object cannot be interpreted as an integer
`

mnist.py execution fails.

Windows 10
Python 3.6.2

Tensorflow version problem

Even after installing python3.6, tensorflow-gpu, keras, and autokeras, when I do
"import autokeras" or "from autokeras.classifier import ImageClassifier", it gives me

ModuleNotFoundError: No module named 'autokeras.classifier'; 'autokeras' is not a package

• OS type and version: Linux Ubuntu 16.04
• Python: 3.6
• autokeras: 0.2.1
• scikit-learn: 0.19.1
• numpy: 1.15.0
• keras: 2.2.2
• scipy: 1.1.0
• tensorflow: 1.10.0
• pytorch: 0.4.0

How can I fix this?

Export Keras model after fit

I want to save the trained model and make it usable for serving. Is it possible? How?

load_image_dataset not found

Following the instructions of https://autokeras.com/start/
I've found this error
Loading load_image_dataset with
from autokeras.image_classifier import load_image_dataset
rise an error " No module named 'autokeras.image_classifier'"

checking the script in the
rise the error from different module "from autokeras.classifier import load_image_dataset"

I've found the function in the module image_supervised.py

torch version

I am trying to run the code on windows, I have torch version 0.4.1, it's showing that requirements.txt requires torch version 0.4.0 which is unavailable. How do I go about this problem?

super() takes at least 1 argument (0 given)

Hi! HaiFeng, I tried to run the demo, but it failed, the code as shown below:

from keras.datasets import mnist
from autokeras.classifier import ImageClassifier

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(x_train.shape + (1,))
x_test = x_test.reshape(x_test.shape + (1,))

clf = ImageClassifier(verbose=True)
clf.fit(x_train, y_train, time_limit=12 * 60 * 60)
clf.final_fit(x_train, y_train, x_test, y_test, retrain=True)
y = clf.evaluate(x_test, y_test)
print(y)

The error message as shown below:

Initializing search.
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-23-077a1b1d90e1> in <module>()
      1 clf = ImageClassifier(verbose=True)
----> 2 clf.fit(x_train, y_train, time_limit=12 * 60 * 60)
      3 clf.final_fit(x_train, y_train, x_test, y_test, retrain=True)
      4 y = clf.evaluate(x_test, y_test)
      5 print(y)

/Users/victor/virtualenvlist/mydlp2/lib/python2.7/site-packages/autokeras/classifier.pyc in fit(self, x_train, y_train, time_limit)
    210         start_time = time.time()
    211         while time.time() - start_time <= time_limit:
--> 212             run_searcher_once(x_train, y_train, x_test, y_test, self.path)
    213             if len(self.load_searcher().history) >= constant.MAX_MODEL_NUM:
    214                 break

/Users/victor/virtualenvlist/mydlp2/lib/python2.7/site-packages/autokeras/classifier.pyc in run_searcher_once(x_train, y_train, x_test, y_test, path)
     41         backend.set_session(sess)
     42     searcher = pickle_from_file(os.path.join(path, 'searcher'))
---> 43     searcher.search(x_train, y_train, x_test, y_test)
     44 
     45 

/Users/victor/virtualenvlist/mydlp2/lib/python2.7/site-packages/autokeras/search.pyc in search(self, x_train, y_train, x_test, y_test)
    156     def search(self, x_train, y_train, x_test, y_test):
    157         if not self.history:
--> 158             self.init_search()
    159 
    160         # Start the new process for training.

/Users/victor/virtualenvlist/mydlp2/lib/python2.7/site-packages/autokeras/search.pyc in init_search(self)
    142             print('Initializing search.')
    143         graph = DefaultClassifierGenerator(self.n_classes,
--> 144                                            self.input_shape).generate(self.default_model_len,
    145                                                                       self.default_model_width)
    146         model_id = self.model_count

/Users/victor/virtualenvlist/mydlp2/lib/python2.7/site-packages/autokeras/generator.pyc in __init__(self, n_classes, input_shape)
     34 class DefaultClassifierGenerator(ClassifierGenerator):
     35     def __init__(self, n_classes, input_shape):
---> 36         super().__init__(n_classes, input_shape)
     37 
     38     def generate(self, model_len=constant.MODEL_LEN, model_width=constant.MODEL_WIDTH):

TypeError: super() takes at least 1 argument (0 given)

From the log, I think the fit() function in class ImageClassifier got something wrong.

Thanks in Advances.

test modue now working on mac

from keras.datasets import mnist
from autokeras.image_supervised import ImageClassifier

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(x_train.shape + (1,))
x_test = x_test.reshape(x_test.shape + (1,))

clf = ImageClassifier(verbose=True)
clf.fit(x_train, y_train, time_limit=12 * 60 * 60)
clf.final_fit(x_train, y_train, x_test, y_test, retrain=True)
y = clf.evaluate(x_test, y_test)
print(y)

Bug Description

Reproducing Steps

Steps to reproduce the behavior:

• Step 1: ...
• Step 2: ...

Expected Behavior

Setup Details

Include the details about the versions of:

• OS type and version: mac high sierra 10.13.6(17G65)
• Python: 3.6(anaconda)
• autokeras:
• scikit-learn: 0.19.1
• numpy: 1.14.5
• keras:2.2.2
• scipy:1.1.0
• tensorflow: 1.8.0
• pytorch: not installed

Additional context

cannot import name Constant

This is my error:

Using TensorFlow backend.
Traceback (most recent call last):
File "autok1.py", line 3, in
from autokeras.constant import Constant
ImportError: cannot import name 'Constant'

This looks like an internal error?

ImageSegmentaion

Task Module: Semantic Segmentation (2D)

Task Description

• Input: one image or more, each of them contains one or more objects
• Output: for each input image, each pixel on the image is classified which object or class it belongs to.

Evaluation Metrics: mIoU (mean intersection over union)

1. For each segmentation generated, the foreground IoU is calculated, which is between 0~1
2. Then IoU is computed on a per-class basis and then averaged as mIoU

Benchmark Datasets

1. PASCAL Visual Object Classification (PASCAL VOC) dataset
2. Common Objects in COntext (COCO) dataset
3. Cityscapes

Reason

Semantic Segmentation is one of the most important research areas and applications in Computer Vision, such as autonomous driving, indoor navigation. This module will provide easy access to people or group who needs to perform semantic segmentation in their work with limited Computer Vision or Machine Learning background.

Solution

Some add-ons on current AutoKeras to be considered:

1. Add unpooling layer into search space, and apply restrictions during search (e.g. alignment of pooling/unpooling layers).
2. Consider using some popular architectures as initializations (e.g. UNet, DeepLab).
3. Consider adding complexity constraint (e.g. depth, width) in searching metrics. Currently the accuracy is the only metrics for the model selection.

Incompatible with Google Colaboratory

Bug Description

As stated here, Autokeras kills the Google Colaboratory kernel, in CPU or GPU mode.

Reproducing Steps

Run the following notebook, which is a simple copy-paste of the AutoKeras tutorial, with a slight modification: in the AutoKeras API, the package autokeras.image_classifier seems to have been renamed to autokeras.image_supervised, and the documentation is not yet up to date.

Expected Behavior

It should run, since Colaboratory runs Python 3.6

Additional context

Colaboratory seems incompatible with the Python multiprocessing API.

I modified the code of search.py/Searcher in order to remove it, and it worked.
More clearly, I replaced this:
multiprocessing.set_start_method('spawn', force=True)
pool = multiprocessing.Pool(1)
train_results = pool.map_async(train, [(graph, train_data, test_data, self.trainer_args, os.path.join(self.path, str(model_id) + '.png'), self.metric, self.loss, self.verbose)])
By this:
train_results = train((graph, train_data, test_data, self.trainer_args, os.path.join(self.path, str(model_id) + '.png'), self.metric, self.loss, self.verbose))

Maybe AutoKeras could have an option in order to desactivate the use of that API ?

ImageClassifier1D

Feature Description

CNN structure is becoming more and more widely used in text classification. Unlike image in shape [height, weight, channel], in this case the input will be in shape [1, n, embedding size], where n is the length of the text. Currently such input will cause error in autokeras, since "1 - filter_size" will be less than zero. So I think autokeras should automatically detect such input shape and then apply 1D architecture search like 1D convolution, 1D pooling, etc.

Reason

Autokeras now is mainly for 2D input like image. It could be more flexible.

Solution

Add some codes to detect the shape of the input, and to search specific network structures optimized for 1D, 2D or 3D tensor depending on the shape of the input.

ImportError: No module named queue

from queue import PriorityQueue

ImportError: No module named queue

Support visual regression

Describe which feature do you want to add to this project
Visual Regression would allow this project to search for an architecture adept at predicting values for images, rather than classifying them into categories.

Explain why this feature is needed
This would allow use cases like the following to work:

1. Self Driving Car - Steering Angle prediction from lane images
2. LaMem (from MIT) - Image Memorability prediction

Describe the solution you'd like
Add an ImageRegressor in a similar style as ImageClassifier.

Describe alternatives you've considered
You could merge the classifier and regressor into one 'Visual' category, allowing other categories (e.g. 'Text', 'Audio') to come later.

Additional context
N/A

can autokeras support distributed mode?

Thanks for Auto Keras.
I have one question.
Since searching model takes long time, I believe it would be awesome if the Autokeras supports trainining multiple models on distributed machines (parallel training). In this sense, searching time can be saved a lot.
I wonder if Auto Keras is already ready for it or have plan to support this?

Thanks!

Parallel on Multiple GPUs

I tried fitting the data on amazon p2.8xlarge instance(which has 8 GPUs), but it just fills the VRAM of one GPU and crashes with the cuda out of memory error. Is there any way of using all 8 GPUs?