r-tensorflow / autokeras

Package: R Interface to AutoKeras

r-tensorflow website

This is a blogdown website and can be previewed locally with

blogdown::serve_site()

It relies on Hugo v0.58.3.

Updating the API reference:

The API reference can be built with:

Rscript /scripts/api.R

Updating the guides

You can update the Guide section with:

Rscript /scripts/guide.R

autokeras's Issues

How to merge a CNN with RNN using Autokeras? It seems the Merge() does not support this. I tried merge_type flags as 'concatenate' or 'add' but non of them works. In particular I want to find an optimum model made of a ConvBlock() followed by a RNNBlock().

Unable to install

tried installing it on 2 separate machines, but I get the same error:

Error in install_keras(method = method, conda = conda, version = keras, : You should call install_keras() only in a fresh R session that has not yet initialized Keras and TensorFlow (this is to avoid DLL in use errors during installation)

Tried a couple of different things, creating new anaconda environments, reinstalling python, installing miniconda, etc.

I can install the keras and tensorflow packages separately w/o much problems, just the autokeras is an issue.

If you have any idea what to do would appreaciate it.

Unable to load Autokeras Package in R

I installed Autokeras using Docker and am attempting to use Autokeras in R, but I am unable to load the package.

I used the following commands to download Autokeras:

In terminal:
docker pull jcrodriguez1989/r-autokeras:0.1.0 docker run -it jcrodriguez1989/r-autokeras:0.1.0 /bin/bash

After doing this I was able to load the Autokeras package in R, but attempting to run the function model_text_classifier resulted in the following error:

Use the install_keras() function to install the core Keras library
Error: Error loading Python module keras

Because of this I tried the Autokeras install function in R:
remotes::install_github("jcrodriguez1989/autokeras", force=TRUE)

Since doing this I receive the following error whenever I attempt library(autokeras)

Error: package or namespace load failed for ‘autokeras’:
 .onLoad failed in loadNamespace() for 'autokeras', details:
  call: py_module_import(module, convert = convert)
  error: ModuleNotFoundError: No module named 'autokeras'

I have also noticed that when I run reticulate::py_config() in R I receive the following output:

python:         /Users/brianmoser/Library/r-miniconda/envs/r-reticulate/bin/python
libpython:      /Users/brianmoser/Library/r-miniconda/envs/r-reticulate/lib/libpython3.8.dylib
pythonhome:     /Users/brianmoser/Library/r-miniconda/envs/r-reticulate:/Users/brianmoser/Library/r-miniconda/envs/r-reticulate
version:        3.8.5 | packaged by conda-forge | (default, Jul 31 2020, 02:18:36)  [Clang 10.0.1 ]
numpy:          /Users/brianmoser/Library/r-miniconda/envs/r-reticulate/lib/python3.8/site-packages/numpy
numpy_version:  1.18.5

Although the version says (default, Jul 31 2020) I did not remember having Python 3.8 installed, and my default Python version when opening Python in the terminal is 3.6.10. I also see that Python 3.6.10 is saved in the base environment at /Users/brianmoser/opt/anaconda3/lib/python3.6, but is not saved in /Users/brianmoser/Libraryr-miniconda/envs/r-reticulate/lib/ folder.

Given all of this information, is there any clear solution to make Autokeras function in R? Is the problem associated with the presence of multiple versions of Python or is it something else?

Additional build info:
OS: macOS High Sierra 10.13.4
R: 3.6.3
RStudio: 1.2.5042

install error

autokeras is now 0.4.0 .

autokeras 0.3.7 is depends tensorflow==1.12.0 but it's removed on pip or conda.

need to upgrade.

Support MXNet Backend

Fix for source code

In model_image_classifier.R, model_image_regressor.R, model_structured_data_classifier.R, model_structured_data_regressor.R, model_text_classifier.R, model_text_regressor.R, I had to change the "name" parameter to "project_name" for it to be accepted by Python autokeras 1.0.12. I made these changes to R autokeras 1.0.1 source code and then packed everything back into a tar.gz which I then installed in R.

ValueError: Unknown layer: IdentityLayer

Hello,

I built my own autokeras model, and it successfully ran and created an autokeras model. Next, I exported and saved the model via the instructions in the vignette:

(keras_model <- export_model(akm1))

# And save the Keras model
keras::save_model_hdf5(keras_model, model_file)

However, when I try and reload the model with my_model <- keras::load_model_hdf5(model_file) I get the following error:

 Error in py_call_impl(callable, dots$args, dots$keywords) : 
     ValueError: Unknown layer: IdentityLayer

Any insight or help would be appreciated!

> keras:::keras_version()
[1] ‘2.2.4’
> tensorflow:::tf_version()
[1] ‘2.0’

Error in py_call_impl(callable, dots$args, dots$keywords) : AttributeError: 'TrackableWeightHandler' object has no attribute 'shape'

Hi,

Thank you for developing the "autokeras" package. It will definitely be a helpful tool for data scientists. I am trying to run the example for model_structured_data_regressor but I get errors with the 'evaluation' and 'export_model'. It appears that autokeras is unable to select the 'best' trained model

Below is the output of the code and my r session info

Any assistance is highly appreciated

Thank you,

Jacques

Predict with the best model

(predicted_y <- reg %>% predict(test_file_to_predict))
[,1]
[1,] 2.564118
[2,] 2.381873
[3,] 2.419431
[4,] 2.604501
[5,] 2.677018
[6,] 2.449875
[7,] 2.924572
[8,] 2.533927
[9,] 2.790744
[10,] 2.510988
[11,] 3.068361
[12,] 2.985615
[13,] 2.941829
[14,] 2.945893
[15,] 2.958850
[16,] 2.731235
[17,] 2.757782
[18,] 3.158206
[19,] 2.636754
[20,] 2.885813
[21,] 3.081780
[22,] 3.282970
[23,] 2.709834
[24,] 3.087875
[25,] 3.195198
[26,] 2.692371
[27,] 2.763576
[28,] 2.984694
[29,] 3.079619
[30,] 3.038662

Evaluate the best model with testing data

reg %>% evaluate(test_file_to_eval, "Sepal.Length")
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.

[[1]]
[1] 9.185607
1/1 [==============================] - 0s 67ms/step - loss: 9.1856 - mean_squared_error: 9.1856

[[2]]
[1] 9.185607

Get the best trained Keras model, to work with the keras R library

export_model(reg)

(keras_model <- export_model(reg))
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.iter
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_1
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.beta_2
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.decay
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:Unresolved object in checkpoint: (root).optimizer.learning_rate
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
WARNING:tensorflow:A checkpoint was restored (e.g. tf.train.Checkpoint.restore or tf.keras.Model.load_weights) but not all checkpointed values were used. See above for specific issues. Use expect_partial() on the load status object, e.g. tf.train.Checkpoint.restore(...).expect_partial(), to silence these warnings, or use assert_consumed() to make the check explicit. See https://www.tensorflow.org/guide/checkpoint#loading_mechanics for details.
Error in py_call_impl(callable, dots$args, dots$keywords) :
AttributeError: 'TrackableWeightHandler' object has no attribute 'shape'

sessionInfo()
R version 3.6.3 (2020-02-29)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.1252 LC_CTYPE=English_United States.1252 LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C LC_TIME=English_United States.1252

attached base packages:
[1] stats graphics grDevices utils datasets methods base

other attached packages:
[1] reticulate_1.14 tfruns_1.4 tensorflow_2.0.0 keras_2.2.5.0 autokeras_1.0.1

loaded via a namespace (and not attached):
[1] Rcpp_1.0.3 zeallot_0.1.0 rappdirs_0.3.1 R6_2.4.1 jsonlite_1.6.1 magrittr_1.5 whisker_0.4 generics_0.0.2 tools_3.6.3
[10] xfun_0.12 yaml_2.2.1 compiler_3.6.3 base64enc_0.1-3 knitr_1.28

How to set CPU numbers to run?

Hi,

Thanks for providing this interface in R. I am wondering if there is an easy way to set CPU to run autokeras?

Best,

Shixiang

dropout rate value for best model

Dear Juan Cruz, I was wondering how to access the dropout rate for dropout layers selected for the best model obtained with the export_model function.
Thanks again for autokeras.