(define-module (gn packages machine-learning)
  #:use-module ((guix licenses) #:prefix license:)
  #:use-module (guix packages)
  #:use-module (guix utils)
  #:use-module (gnu packages machine-learning)
  #:use-module (guix download)
  #:use-module (guix build-system python)
  #:use-module (gnu packages python-xyz))

(define-public tensorflow-native
  (package
    (inherit tensorflow)
    (name "tensorflow-native")
    (arguments
     (substitute-keyword-arguments (package-arguments tensorflow)
       ((#:substitutable? _ #f) #f)
       ((#:configure-flags flags)
        `(cons
           "-Dtensorflow_OPTIMIZE_FOR_NATIVE_ARCH=ON"
           (delete "-Dtensorflow_OPTIMIZE_FOR_NATIVE_ARCH=OFF"
                   ,flags)))))))

(define-public tensowflow-native-instead-of-tensorflow
  (package-input-rewriting/spec `(("tensorflow" . ,(const tensorflow-native)))))



(define-public python-keras-preprocessing
  (package
    (name "python-keras-preprocessing")
    (version "1.1.0")
    (source
     (origin
       (method url-fetch)
       (uri (pypi-uri "Keras_Preprocessing" version))
       (sha256
        (base32
         "1r98nm4k1svsqjyaqkfk23i31bl1kcfcyp7094yyj3c43phfp3as"))))
    (build-system python-build-system)
    (propagated-inputs
     (list python-numpy python-six))
    (native-inputs
     (list python-pandas
           python-pillow
           python-pytest
           python-pytest-cov
           python-pytest-xdist
           tensorflow))
    (home-page "https://github.com/keras-team/keras-preprocessing/")
    (synopsis "Data preprocessing and augmentation for deep learning models")
    (description
     "Keras Preprocessing is the data preprocessing and data augmentation
module of the Keras deep learning library.  It provides utilities for working
with image data, text data, and sequence data.")
    (license license:expat)))


(define-public python-keras-no-tests
  (package
    (name "python-keras-no-tests")
    (version "2.3.1")
    (source
     (origin
       (method url-fetch)
       (uri (pypi-uri "Keras" version))
       (sha256
        (base32
         "1k68xd8n2y9ldijggjc8nn4d6d1axw0p98gfb0fmm8h641vl679j"))
       (modules '((guix build utils)))
       (snippet
        '(substitute* '("keras/callbacks/callbacks.py"
                        "keras/engine/training_utils.py"
                        "keras/engine/training.py"
                        "keras/engine/training_generator.py"
                        "keras/utils/generic_utils.py")
           (("from collections import Iterable")
            "from collections.abc import Iterable")
           (("collections.Container")
            "collections.abc.Container")
           (("collections.Mapping")
            "collections.abc.Mapping")
           (("collections.Sequence")
            "collections.abc.Sequence")))))
    (build-system python-build-system)
    (arguments
     `(#:phases
       (modify-phases %standard-phases
         (add-after 'unpack 'tf-compatibility
           (lambda _
             (substitute* "keras/backend/tensorflow_backend.py"
               (("^get_graph = .*")
                "get_graph = tf.get_default_graph")
               (("tf.compat.v1.nn.fused_batch_norm")
                "tf.nn.fused_batch_norm")
               ;; categorical_crossentropy does not support axis
               (("from_logits=from_logits, axis=axis")
                "from_logits=from_logits")
               ;; dropout accepts a level number, not a named rate argument.
               (("dropout\\(x, rate=level,")
                "dropout(x, level,")
               (("return x.shape.rank")
                "return len(x.shape)"))))
         (add-after 'unpack 'hdf5-compatibility
           (lambda _
             ;; The truth value of an array with more than one element is ambiguous.
             (substitute* "tests/keras/utils/io_utils_test.py"
               ((" *assert .* == \\[b'(asd|efg).*") ""))
             (substitute* "tests/test_model_saving.py"
               (("h5py.File\\('does not matter',")
                "h5py.File('does not matter', 'w',"))
             (substitute* "keras/utils/io_utils.py"
               (("h5py.File\\('in-memory-h5py', driver='core', backing_store=False\\)")
                "h5py.File('in-memory-h5py', 'w', driver='core', backing_store=False)")
               (("h5file.fid.get_file_image")
                "h5file.id.get_file_image"))
             (substitute* "keras/engine/saving.py"
               (("\\.decode\\('utf-?8'\\)") ""))))
         (add-after 'unpack 'delete-unavailable-backends
           (lambda _
             (delete-file "keras/backend/theano_backend.py")
             (delete-file "keras/backend/cntk_backend.py")))
         (delete 'check))))
    (propagated-inputs
     (list python-h5py
           python-keras-applications
           python-keras-preprocessing
           python-numpy
           python-pydot
           python-pyyaml
           python-scipy
           python-six
           tensorflow
           graphviz))
    (native-inputs
     (list python-flaky
           python-markdown
           python-pandas
           python-pytest
           python-pytest-cov
           python-pytest-timeout
           python-pytest-xdist
           python-pyux
           python-sphinx
           python-requests))
    (home-page "https://keras.io/")
    (synopsis "High-level deep learning framework")
    (description "Keras is a high-level neural networks API, written in Python
and capable of running on top of TensorFlow.  It was developed with a focus on
enabling fast experimentation.  Use Keras if you need a deep learning library
that:
@itemize
@item Allows for easy and fast prototyping (through user friendliness,
  modularity, and extensibility).
@item Supports both convolutional networks and recurrent networks, as well as
  combinations of the two.
@item Runs seamlessly on CPU and GPU.
@end itemize\n")
    (license license:expat)))