Develop

.github/workflows/build-cpu-windows.yml

@@ -15,13 +15,13 @@ jobs:
       run: |
         mkdir build
         cd build
-        cmake -DBUILD_SUPERBUILD=ON -DBUILD_SHARED_LIBS=OFF ..
+        cmake -G "Visual Studio 16 2019" -A x64 -T ClangCL -DBUILD_SUPERBUILD=ON -DBUILD_SHARED_LIBS=OFF -DBUILD_TARGET=CPU ..
       shell: cmd
     - name: Build
       run: cmake --build build --config Release
       shell: cmd
     - name: Test
       run: |
-            cd build
-            "bin/Release/unit_tests.exe"
+        cd build
+        "bin/Release/unit_tests.exe"
       shell: cmd

.gitignore

@@ -21,7 +21,6 @@ temp/
 docs/doxygen/*
 !docs/doxygen/Doxyfile
 docs/sphinx/source/_build
-docs/sphinx/source/_static
 
 # Other files
 *.dot
@@ -32,8 +31,8 @@ old_build/example_*
 old_build/test_*
 
 # ONNX
-#onnx.pb.h
-#onnx.pb.cc
+onnx.pb.h
+onnx.pb.cc
 
 # Files generated by CLion
 cmake-build-*

cmake/CMakeLists.txt

@@ -159,7 +159,7 @@ else()
 
     set(Protobuf_ROOT "${EP_BASE_DIR}/protobuf" PARENT_SCOPE)
     set(Protobuf_INCLUDE_DIRS "${EP_BASE_DIR}/protobuf/include" PARENT_SCOPE)
-    set(Protobuf_LIBRARIES_DIRS "${EP_BASE_DIR}/protobuf/lib")
+    set(Protobuf_LIBRARIES_DIRS "${EP_BASE_DIR}/protobuf/lib")  # TODO: Fix. Sometimes is /lib, and others /lib64
     set(Protobuf_LIBRARIES_DIRS ${Protobuf_LIBRARIES_DIRS} PARENT_SCOPE)
     # set(Protobuf_LIBRARIES "protobuf" PARENT_SCOPE)
     set(Protobuf_PROTOC_EXECUTABLE "${EP_BASE_DIR}/protobuf/bin/protoc" PARENT_SCOPE)

docs/markdown/development.md

@@ -79,6 +79,23 @@ Drop the `formulas/brew/eddl.rb` file at `/usr/local/Homebrew/Library/Taps/homeb
 Then you can also execute; `brew`
 
 
+## Create conda env
+
+```
+# [YAML] Create environment
+conda env create -f environment.yml
+
+# [RAW] Create environment
+conda create --name myenv
+
+# Update environment
+conda env update -f environment.yml
+
+# Remove environment
+conda remove --name eddl --all
+```
+
+
 ## Build conda package
 
 ```

docs/markdown/eddl_progress_tensor.md

@@ -58,10 +58,7 @@ Numpy-like operations over a raw-tensor object
 | ------------- | ---- | ---- | -------- |
 | randu        | ✔️    | ✔️     | 	Return a uniform random matrix with given shape.   |
 | randn        | ✔️    | ✔️    | 	Return a normal random matrix with data from the "standard normal" distribution.     |
-| rand_uniform        | ✔️    | ✔️    |     |
-| rand_signed_uniform        | ✔️    | ✔️    |     |
-| rand_normal        | ✔️    | ✔️    |     |
-| rand_binary        | ✔️    | ✔️    |     |
+
 
 
 ### Building matrices
@@ -99,6 +96,17 @@ Numpy-like operations over a raw-tensor object
 | unsqueeze      | ✔️    | ✔️    |  Expand the shape of an array.  `[3, 4, 7] => [1, 3, 4, 7]`      |
 
 
+### Value operations
+
+| Functionality | CPU  | GPU  | Comments |
+| ------------- | ---- | ---- | -------- |
+| fill        | ✔️    | ✔️    |  Fills a tensor in-place, with a constant value   |
+| fill_rand_uniform        | ✔️    | ✔️    |  Fills a tensor in-place, with values randomly sampled from a uniform distribution   |
+| fill_rand_signed_uniform        | ✔️    | ✔️    | Fills a tensor in-place, with values randomly sampled from a signed uniform distribution    |
+| fill_rand_normal        | ✔️    | ✔️    |  Fills a tensor in-place, with values randomly sampled from a normal distribution   |
+| fill_rand_binary        | ✔️    | ✔️    | Fills a tensor in-place, with values randomly sampled from a binary distribution    |
+
+
 ### Joining arrays
 
 | Functionality | CPU  | GPU  | Comments |

docs/sphinx/source/_templates/searchbox.html

@@ -0,0 +1,14 @@
+<!--<div>-->
+<!--    <h3>{{ _('Quick search') }}</h3>-->
+<!--    <script>-->
+<!--        (function() {-->
+<!--            var cx = 'aaa06ecc8743366d5';-->
+<!--            var gcse = document.createElement('script');-->
+<!--            gcse.async = true;-->
+<!--            gcse.src = 'https://cse.google.com/cse.js?cx=' + cx;-->
+<!--            var s = document.getElementsByTagName('script')[0];-->
+<!--            s.parentNode.insertBefore(gcse, s);-->
+<!--        })();-->
+<!--    </script>-->
+<!--    <gcse:search></gcse:search>-->
+<!--</div>-->

docs/sphinx/source/bundle/computing_service.rst

@@ -22,8 +22,14 @@ Example:
 GPU
 ====
 
-.. doxygenfunction:: eddl::CS_GPU(const vector<int>, int, string)
-.. doxygenfunction:: eddl::CS_GPU(const vector<int>, string)
+.. doxygenfunction:: eddl::CS_GPU(const vector<int> g)
+
+.. doxygenfunction:: eddl::CS_GPU(const vector<int> g, int lsb)
+
+.. doxygenfunction:: eddl::CS_GPU(const vector<int> g, int lsb, string mem)
+
+.. doxygenfunction:: eddl::CS_GPU(const vector<int> g, string mem)
+
 
 Example:
 

docs/sphinx/source/bundle/initializers.rst

@@ -58,18 +58,6 @@ Example:
 
     layer RandomUniform(layer l, float min=0.0,float max=0.1, float seed=1234);
 
-HeUniform
--------------
-
-.. doxygenfunction:: HeUniform
-
-Example:
-
-.. code-block:: c++
-   :linenos:
-
-    layer HeUniform(layer l,int seed=1234);
-
 
 Constant
 -------------

docs/sphinx/source/conf.py

@@ -19,8 +19,8 @@
 # -- Project information -----------------------------------------------------
 
 project = 'EDDL'
-copyright = '2020, eddl'
-author = 'Pattern Recognition and Human Language Translation Research Center.'
+copyright = '2020, EDDL'
+author = 'Pattern Recognition and Human Language Technology research center'
 
 
 # -- General configuration ---------------------------------------------------
@@ -31,6 +31,8 @@
 extensions = [
     'breathe',
     'sphinx_tabs.tabs',
+    # 'sphinxprettysearchresults',
+    # 'sphinxcontrib.lunrsearch',
 
     'sphinx.ext.autodoc',
     'sphinx.ext.doctest',
@@ -45,7 +47,7 @@
 breathe_default_project = "eddl"
 
 # Add any paths that contain templates here, relative to this directory.
-templates_path = ['_templates']
+# templates_path = ['_templates']
 
 # List of patterns, relative to source directory, that match files and
 # directories to ignore when looking for source files.
@@ -68,3 +70,7 @@
 html_static_path = ['_static']
 
 pygments_style = 'sphinx'
+
+# html_sidebars = {
+#     '**': ['searchbox.html'],
+# }

docs/sphinx/source/index.rst

@@ -105,27 +105,18 @@ The code is open source, and `available on github`_.
     datasets/segmentation.rst
 
 
-.. toctree::
-    :maxdepth: 1
-    :caption: Miscellaneous
-
-    misc/misc.rst
-
-
 .. toctree::
     :maxdepth: 1
     :caption: Tensor
 
     tensor/create.rst
     tensor/manipulation.rst
-    tensor/binary.rst
+    tensor/image.rst
     tensor/indexing.rst
     tensor/input_output.rst
     tensor/linear_algebra.rst
     tensor/logic_functions.rst
-    tensor/mask_ops.rst
     tensor/math.rst
-    tensor/misc.rst
 
 
 Indices and tables

docs/sphinx/source/installation/build-options.rst

@@ -22,27 +22,33 @@ environment by running the following commands **from the source directory**:
     conda env create -f environment.yml
     conda activate eddl
 
+You can also update your environment with:
+
+.. code:: bash
+
+    conda env update -f environment.yml
+
 If you decide to manually install these dependencies in your system (make sure they are at standard paths):
 
 .. code::
 
     - cmake>=3.9.2
     - eigen>=3.3.7
-    - zlib=1.2.*
-    - protobuf=3.11.*
+    - protobuf
+    - libprotobuf  # We need to avoid problems with paths (idk why)
+    - zlib
     - cudatoolkit
     - gtest
     - graphviz  # Build & Run
     - wget
     - doxygen  # Docs
     - python
     - pip
-
     - pip:
-        - sphinx==3.0.3
-        - sphinx_rtd_theme==0.4.3
-        - sphinx-tabs==1.1.13
-        - breathe==4.16.0
+    - sphinx==3.2.1
+    - sphinx_rtd_theme==0.5.0
+    - sphinx-tabs==1.3.0
+    - breathe==4.22.1
 
 
 .. note::
@@ -125,6 +131,10 @@ troubleshoot the compilation process (see: :doc:``troubleshoot``).
 
     -DCMAKE_PREFIX_PATH=/path/to/dir
 
+.. note::
+
+    If using conda, get the path by activating the environment, and typing ``echo $CONDA_PREFIX``
+
 
 - **Installation paths:** To change the installation paths, use the following cmake option:
 
@@ -135,13 +145,14 @@ troubleshoot the compilation process (see: :doc:``troubleshoot``).
 .. note::
 
     Defaults to ``/usr/local`` on UNIX and ``c:/Program Files`` on Windows.
+    If using conda, get the path by activating the environment, and typing ``echo $CONDA_PREFIX``
 
 
 - **C++ compiler:** If you have problems with the default g++ compiler, try setting ``EIGEN3_INCLUDE_DIR``, such as:
 
 .. code:: bash
 
-    -DCMAKE_CXX_COMPILER=/path/to/c++compiler
+    -DCMAKE_CXX_COMPILER=/path/to/c++compiler  # /usr/bin/g++-8
 
 .. note::
 
@@ -152,25 +163,35 @@ troubleshoot the compilation process (see: :doc:``troubleshoot``).
 
 .. code:: bash
 
-    -DCMAKE_CUDA_COMPILER=/path/to/cuda compiler
+    -DCMAKE_CUDA_COMPILER=/path/to/cuda compiler  #/usr/bin/nvcc
 
 .. note::
 
     You can also create a symbolic link: (unix) ``sudo ln -s usr/local/cuda-{VERSION} /usr/local/cuda``
 
 
+- **CUDA host compiler:** If cmake have problems finding your cuda host compiler, try setting ``CMAKE_CUDA_COMPILER``, such as:
+
+.. code:: bash
+
+    -DCMAKE_CUDA_HOST_COMPILER=/path/to/cuda compiler  # /usr/bin/g++-8
+
+.. note::
+
+    You can also create a symbolic link: (unix) ``sudo ln -s usr/local/cuda-{VERSION} /usr/local/cuda``
+
 - **CUDA Toolkit:** If CMake is unable to find CUDA automatically, try setting ``CUDA_TOOLKIT_ROOT_DIR``, such as:
 
 .. code:: bash
 
-    -DCUDA_TOOLKIT_ROOT_DIR=/path/to/cuda
+    -DCUDA_TOOLKIT_ROOT_DIR=/path/to/cuda   # /usr/local/cuda-11.1
 
 
 - **Eigen3:** At the core of many numerical operations, we use Eigen3_. If CMake is unable to find Eigen3 automatically, try setting ``Eigen3_DIR``, such as:
 
 .. code:: bash
 
-    -DEigen3_DIR=/path/to/eigen
+    -DEigen3_DIR=/path/to/eigen  # /usr/lib/cmake/eigen3
 
 
 - **Use OpenMP:** To enable/disabled OpenMP, use the setting ``BUILD_OPENMP``, such as:

docs/sphinx/source/installation/faq.rst

@@ -2,42 +2,47 @@ FAQ
 ===
 
 
+Python library
+---------------
+
 Is there a Python version?
---------------------------
+^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 Yes, the PyEDDL_ is the EDDL version for the Python lovers
 
 
-Can I contribute?
-------------------
+Is the PyEDDL develop by the same team
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-Yes, but first open a new issue to explain and discuss your contribution.
+Sort of. We all are part of the DeepHealth project, and although each team is specialized
+in developing and supporting different libraries, we all contribute to the other projects so that all the DeepHealth
+ecosystem can work smoothly.
 
 
-Can I control the memory consumption?
--------------------------------------
+Contributions
+---------------
 
-Yes, we offer several memory levels to control the memory-speed trade-off. These levels are:
 
+Can I contribute?
+^^^^^^^^^^^^^^^^^^
+
+Yes, but first open a new issue to explain and discuss your contribution.
 
-- ``full_mem`` (default): No memory bound (highest speed at the expense of the memory requirements)
-- ``mid_mem``: Slight memory optimization (good trade-off memory-speed)
-- ``low_mem``: Optimized for hardware with restricted memory capabilities.
 
+Performance
+---------------
 
 Is it faster than PyTorch/TensorFlow/etc
-----------------------------------------
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-Depends... Generally, there are many nuances so that a high-performance can be achieved (see benchmark section) and many of
-those nuances are strongly tied to the classical memory-speed trade-off. To sum up, matrix multiplications on the CPU
-are a 300% faster than their TensorFlow counterpart. Convolution times on GPU (no cuDNN) are slightly faster than on the
-EDDL than PyTorch (210ms vs. 246ms), and using cuDNN  will get the cuDNN performance.
+Check our benchmakrs: `EDDL benchmarks`_
 
 
 Is it more memory-efficient than PyTorch/TensorFlow/etc
--------------------------------------------------------
-
-Depends on your memory setting (see the "Can I control the memory consumption?" answer).
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
+Depends on your memory setting, see the "Can I control the memory consumption?" answer.
+Also, you can take a look at our benchmakrs: `EDDL benchmarks`_
 
-.. _PyEDDL: https://github.com/deephealthproject/pyeddl
+.. _PyEDDL: https://github.com/deephealthproject/pyeddl
+.. _`EDDL benchmarks`: https://github.com/jofuelo/eddl_benchmark