tensorflow-ruby 0.2.0

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA256:
+  metadata.gz: 29ce3a9ced9660aa6b78013f05d631397a41fd060dc22485c170f1f3501e0ce8
+  data.tar.gz: 92130ad0bd4628b2604409c7f4c215fc645233f27b36bea4fc1d00efedd06959
+SHA512:
+  metadata.gz: 3649ed6ff6b7e1f5ab24387ea864392d0956bb0991549b7ced5c76ca224bd9fd82b8b82e4e34aa129e09a04ad3eb775f9db61a983be202cd65f09afe311e0b3c
+  data.tar.gz: 165b341e62aec2ce1c82ba0bb1105e8061bf5d138f8ccb757f04ae5584841fef1d9056d48bb4e398ae965e11fa905db4376ba1909214ae7031ea68d8f33ec05e

data/CHANGELOG.md

@@ -0,0 +1,18 @@
+0.2.0 (2019-12-02)
+====================
+* Implementation of gradient descent
+* Support for training models
+* Linear regression example
+* MNIST example
+* Tensorboard support
+
+0.1.0 (2019-11-04)
+====================
+* Ability to call all TensorFlow operations
+* Eager execution mode
+* Graph execution mode
+* Importing and exporting computation graphs
+* Creating and editing computation graphs
+* Creating custom operators
+* Gradients
+

data/LICENSE.txt

@@ -0,0 +1,22 @@
+Copyright (c) 2019 Andrew Kane, Charlie Savage
+
+MIT License
+
+Permission is hereby granted, free of charge, to any person obtaining
+a copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sublicense, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice shall be
+included in all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

data/README.md

@@ -0,0 +1,104 @@
+# Tensorflow
+
+Ruby bindings for [Tensorflow](https://github.com/tensorflow/tensorflow).
+
+## Installation
+
+First you'll need to install Tensorflow 2. You can either do a full [installation](https://www.tensorflow.org/install) or just install [Tensorflow for C](https://www.tensorflow.org/install/lang_c). In both cases, you'll need to make 
+sure the tensorflow library is on the system PATH so the Ruby bindings can load it.
+
+Next install the gem:
+
+```ruby
+gem install 'tensorflow-ruby'
+```
+
+## Overview
+
+The Ruby bindings are on built on top of TensorFlow's [C API] (https://github.com/tensorflow/docs/blob/master/site/en/r1/guide/extend/bindings.md)
+which provides access to some, but not all, of TensorFlow's functionality. The Ruby bindings  
+are designed to mimic the [Python API](https://www.tensorflow.org/versions/r2.0/api_docs/python/tf). Just like the
+Python bindings, Ruby defaults to [eager](https://www.tensorflow.org/guide/eager) execution mode.
+ 
+## Current Status
+
+Feature                                        |  Status
+:--------------------------------------------- | :---------------------------------------------------------- |
+Run a predefined Graph                         | Implemented                                                 |
+Graph construction with generated op functions | Implemented                                                 |
+Gradients                                      | Implemented                                                 |
+Functions                                      | Implemented                                                 |
+Trainers                                       | Gradient Descent Implemented                                |
+Data API                                       | 80% Implemented                                             |
+Tensorboard Support                            | Implemented                                                 |
+Control Flow                                   | Not Implemented                                             |
+Gradient Tape                                  | Not Implemented                                             |
+Keras                                          | Not Implemented                                             |
+
+## Getting Help
+To learn how to use the library:
+* Read this file
+* Take a look at the comprehensive test suite - it will show you examples of using all the binding's functionality
+* Review the gem documentation
+* Review the [Python API](https://www.tensorflow.org/versions/r2.0/api_docs/python/tf) documentation
+* Take a look at the models directory for example implementations (linear regression, mnist neural network, etc.)
+
+## Using tensorflow-ruby
+
+### Constants
+Constants are 
+
+```ruby
+a = Tf.constant([1, 2, 3])
+b = Tf.constant([4, 5, 6])
+a + b
+```
+
+### Variables
+
+```ruby
+v = Tf::Variable.new(0.0)
+w = v + 1
+```
+
+### Math
+
+```ruby
+Tf::Math.abs([-1, -2])
+Tf::Math.sqrt([1.0, 4.0, 9.0])
+```
+
+### Data::Dataset
+
+```ruby
+# load
+train_dataset = Tf::Data::Dataset.from_tensor_slices([train_examples, train_labels])
+test_dataset = Tf::Data::Dataset.from_tensor_slices([test_examples, test_labels])
+
+# shuffle and batch
+train_dataset = train_dataset.shuffle(100).batch(32)
+test_dataset = test_dataset.batch(32)
+
+# iterate
+train_dataset.each do |examples, labels|
+  # ...
+end
+```
+
+## Contributing
+
+Everyone is encouraged to help improve this project. Here are a few ways you can help:
+
+- [Report bugs](https://github.com/cfis/tensorflow-ruby/issues)
+- Fix bugs and [submit pull requests](https://github.com/cfis/tensorflow-ruby/pulls)
+- Write, clarify, or fix documentation
+- Suggest or add new features
+
+To get started with development and testing:
+
+```sh
+git clone https://github.com/cfis/tensorflow-ruby.git
+cd tensorflow-ruby
+bundle install
+rake test
+```

data/lib/datasets/download_manager.rb

@@ -0,0 +1,49 @@
+require 'net/http'
+require 'tmpdir'
+require_relative './resource'
+
+module Tensorflow
+  module Datasets
+    class DownloadManager
+      attr_reader :uri, :dir
+
+      def initialize(dir = Dir.tmpdir)
+        @dir = dir
+      end
+
+      def download(urls)
+        resources = Array(urls).flatten.map {|url| Resource.new(url)}
+        resources.each do |resource|
+          self.download_resource(resource)
+        end
+        resources
+      end
+
+      def download_resource(resource)
+        resource.path = File.join(self.dir, resource.filename)
+        return if File.exist?(resource.path)
+
+        STDOUT << "Downloading #{resource.uri}" << "\n"
+        http = Net::HTTP.new(resource.uri.host, resource.uri.port)
+        http.use_ssl = resource.uri.is_a?(URI::HTTPS)
+
+        request = Net::HTTP::Get.new(resource.uri.request_uri)
+
+        http.start do |http|
+          http.request(request) do |response|
+            file_size = response['content-length'].to_f
+            bytes = 0
+
+            File.open(resource.path, 'wb') do |file|
+              response.read_body do |chunk|
+                file.write(chunk)
+                bytes += chunk.size
+                #STDOUT << "#{(bytes / file_size * 100).to_i}%\r"
+              end
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/datasets/images/mnist.rb

@@ -0,0 +1,54 @@
+$LOAD_PATH.unshift(File.expand_path('../../../lib', __dir__))
+
+require 'tensorflow'
+
+require_relative '../download_manager'
+
+module Tensorflow
+  module Datasets
+    module Images
+      class Mnist
+        extend Decorator
+
+        BASE_URL = 'https://storage.googleapis.com/cvdf-datasets/mnist'
+
+        @tf.function(input_signature=[[:string]])
+        def decode_image(image)
+          image = IO.decode_raw(image, Tf.uint8)
+          image = Tf.cast(image, Tf.float32)
+          image = Tf.reshape(image, [784])
+          # Normalize from [0, 255] to [0.0, 1.0]
+          image / 255.0
+        end
+
+        @tf.function(input_signature=[[:string]])
+        def decode_label(label)
+          # tf.string -> [Tf.uint8]
+          label = Tf::IO.decode_raw(label, Tf.uint8)
+          label = Tf.reshape(label, [])  # label is a scalar
+          Tf.cast(label, Tf.int32)
+        end
+
+        def dataset(images_file, labels_file)
+          download_manager = Datasets::DownloadManager.new
+          urls = ["#{BASE_URL}/#{images_file}",
+                  "#{BASE_URL}/#{labels_file}"]
+
+          resources = download_manager.download(urls)
+
+          images = Data::FixedLengthRecordDataset.new(resources.first.path, 28 * 28, header_bytes: 16, compression_type: 'GZIP').map_func(self.decode_image)
+          labels = Data::FixedLengthRecordDataset.new(resources.last.path, 1, header_bytes: 8, compression_type: 'GZIP').map_func(self.decode_label)
+          Data::ZipDataset.new(images, labels)
+        end
+
+        def train
+          dataset('train-images-idx3-ubyte.gz', 'train-labels-idx1-ubyte.gz')
+        end
+
+        def test
+          dataset('t10k-images-idx3-ubyte.gz', 't10k-labels-idx1-ubyte.gz')
+        end
+      end
+    end
+  end
+end

data/lib/datasets/resource.rb

@@ -0,0 +1,19 @@
+require 'uri'
+
+module Tensorflow
+  module Datasets
+    class Resource
+      attr_reader :uri
+      attr_accessor :path
+
+      def initialize(url)
+        @uri = url.is_a?(URI) ? url : URI.parse(url)
+      end
+
+      def filename
+        File.basename(self.uri.path)
+      end
+    end
+  end
+end
+

data/lib/tensorflow-ruby.rb

@@ -0,0 +1,182 @@
+# dependencies
+require "ffi"
+require "npy"
+require "numo/narray"
+
+# stdlib
+require "digest"
+require "fileutils"
+require "forwardable"
+require "json"
+require "net/http"
+require "tempfile"
+require "zlib"
+
+# Load protobufs. Is this a good idea to require all ruby files in those directories?
+Dir[File.join(__dir__, 'tensorflow', 'core', 'lib', 'core', '*.rb')].each { |file| require file }
+Dir[File.join(__dir__, 'tensorflow', 'core', 'stream_executor', '*.rb')].each { |file| require file }
+Dir[File.join(__dir__, 'tensorflow', 'core', 'framework', '*.rb')].each { |file| require file }
+#Dir[File.join(__dir__, 'tensorflow', 'core', 'protobuf', '*.rb')].each { |file| require file }
+Dir[File.join(__dir__, 'tensorflow', 'core', 'util', '*.rb')].each { |file| require file }
+
+# Ops
+require "tensorflow/ops/audio"
+require "tensorflow/ops/bitwise"
+require "tensorflow/ops/control"
+require "tensorflow/ops/image"
+require "tensorflow/ops/io"
+require "tensorflow/ops/linalg"
+require "tensorflow/ops/math"
+require "tensorflow/ops/nn"
+require "tensorflow/ops/operators"
+require "tensorflow/ops/ops"
+require "tensorflow/ops/random"
+require "tensorflow/ops/raw_ops"
+
+# Core
+require "tensorflow/ffi"
+require "tensorflow/decorators"
+require "tensorflow/error"
+require "tensorflow/execution_context"
+require "tensorflow/name_scope"
+require "tensorflow/op_def_builder"
+require "tensorflow/status"
+require "tensorflow/strings"
+require "tensorflow/tensor_mixin"
+require "tensorflow/tensor_data"
+require "tensorflow/tensor"
+require "tensorflow/variable"
+require "tensorflow/version"
+
+# Extensions
+require "tensorflow/extensions/arg_def.rb"
+require "tensorflow/extensions/boolean.rb"
+require "tensorflow/extensions/narray.rb"
+
+# Printers
+require "tensorflow/printers/graph"
+
+# Eager
+require "tensorflow/eager/context"
+require "tensorflow/eager/operation"
+require "tensorflow/eager/tensor_handle"
+
+# graph
+require "tensorflow/graph/function"
+require "tensorflow/graph/function_def"
+require "tensorflow/graph/gradients"
+require "tensorflow/graph/graph"
+require "tensorflow/graph/graph_def_options"
+require "tensorflow/graph/graph_keys"
+require "tensorflow/graph/operation"
+require "tensorflow/graph/operation_output"
+require "tensorflow/graph/operation_attr"
+require "tensorflow/graph/operation_description"
+require "tensorflow/graph/session"
+
+# Ugly - now require the op gradients
+require "tensorflow/ops/gradients"
+
+
+# specs
+require "tensorflow/type_spec"
+require "tensorflow/batchable_type_spec"
+require "tensorflow/tensor_spec"
+
+# Training
+require "tensorflow/train/optimizer"
+require "tensorflow/train/gradient_descent_optimizer"
+
+# Data
+require "tensorflow/data/dataset"
+require "tensorflow/data/batch_dataset"
+require "tensorflow/data/fixed_length_record_dataset"
+require "tensorflow/data/iterator"
+require "tensorflow/data/map_dataset"
+require "tensorflow/data/repeat_dataset"
+require "tensorflow/data/shuffle_dataset"
+require "tensorflow/data/tensor_dataset"
+require "tensorflow/data/tensor_slice_dataset"
+require "tensorflow/data/tf_record_dataset"
+require "tensorflow/data/zip_dataset"
+
+# Tensorboard
+require "tensorflow/summary"
+require "tensorflow/resource_summary_writer"
+
+# keras
+require "tensorflow/keras/utils"
+require "tensorflow/keras/datasets/boston_housing"
+require "tensorflow/keras/datasets/cifar10"
+require "tensorflow/keras/datasets/cifar100"
+require "tensorflow/keras/datasets/fashion_mnist"
+require "tensorflow/keras/datasets/imdb"
+require "tensorflow/keras/datasets/mnist"
+require "tensorflow/keras/datasets/reuters"
+require "tensorflow/keras/layers/conv"
+require "tensorflow/keras/layers/conv2d"
+require "tensorflow/keras/layers/dense"
+require "tensorflow/keras/layers/dropout"
+require "tensorflow/keras/layers/flatten"
+require "tensorflow/keras/losses/sparse_categorical_crossentropy"
+require "tensorflow/keras/metrics/mean"
+require "tensorflow/keras/metrics/sparse_categorical_accuracy"
+require "tensorflow/keras/model"
+require "tensorflow/keras/models/sequential"
+require "tensorflow/keras/optimizers/adam"
+require "tensorflow/keras/preprocessing/image"
+
+require 'tensorflow/python_compatiblity'
+
+module Tensorflow
+  extend Ops
+
+  GRAPH_MODE = 0
+  EAGER_MODE = 1
+
+  class << self
+    attr_accessor :ffi_lib
+
+    def library_version
+      FFI.TF_Version
+    end
+  end
+
+  # friendlier error message
+  autoload :FFI, "tensorflow/ffi"
+
+  def self.op_defs
+    buffer = FFI.TF_GetAllOpList
+    string = buffer[:data].read_string(buffer[:length])
+    ops = OpList.decode(string)
+    ops.op.each_with_object(Hash.new) do |op_def, hash|
+      hash[op_def.name] = op_def
+    end
+  ensure
+    FFI.TF_DeleteBuffer(buffer)
+  end
+
+  def self.op_def(op_name)
+    self.op_defs[op_name]
+  end
+
+  def self.execution_mode
+    @mode ||= Tensorflow::EAGER_MODE
+  end
+
+  def self.execution_mode=(value)
+    @mode = value
+  end
+
+  def self.name_scope(base_name, &block)
+    ExecutionContext.current.name_scope(base_name, &block)
+  end
+
+  extend PythonCompatability
+  class << self
+    extend Forwardable
+    def_delegators Linalg, *Linalg.singleton_methods
+    def_delegators Math, *Math.singleton_methods
+    def_delegators NN, *NN.singleton_methods
+  end
+end