tensor_stream 0.9.1 → 0.9.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
-SHA1:
-  metadata.gz: 9c6649e7cb423246d4886776bf93a8422e3be936
-  data.tar.gz: c62ffe26455cff685925a5a9a27c49babae6d17e
+SHA256:
+  metadata.gz: 9601653d86556739c89b591768e9d54d13d6335dd2f953fff7d91f22636e8c7b
+  data.tar.gz: 3db4119c9e752df77cbbf7f8f753050a012b03fd2bde04dd6bbfd01f8903af62
 SHA512:
-  metadata.gz: 9dd8519a7c4c34fc0d0c8f4900aa2bec73e6a0dfefe15db4f4c6703188e55080cd2dcce4c8df02993beaa1d6f753b2309fabda623f5ab6a0218c3bd4ce382c0c
-  data.tar.gz: 2e2f051ada324f4f220db702762a179f1ee20985ecbde51d5c8fc976ad50ff32a315459b97a2366c7cf9a718672b42910466b8e1e81ea388a878035807bae1e0
+  metadata.gz: 3ca3eec8ce6cc7e73a1d5e38ccaaf2f65f7a37bb84e0118766de05ec374557847a88f2f14b1e35b896010d7e68a0234589fcf3de95a9417840e2e77ab4a4ea58
+  data.tar.gz: 6b109a2ee9b59e286a2d06d1c3166fc5bbda9a888f0c26aa07bc4f994753ce5dbd7cb0aebb433a7446ddef653f3096ae1b60dbc4610e0dc18c9aaca8d3175f2f

data/CHANGELOG.md

@@ -4,9 +4,14 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/)
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.9.2] - 2018-10-19
+- Add profiling support
+- Make sure sparse ruby arrays are caught
+- Enhancements to support multi-gpu execution
+
 ## [0.9.1] - 2018-10-19
 - Bug fix release
-  
+
 ## [0.9.0] - 2018-10-05
 - Bug fix release for OpenCL gem
 
@@ -16,7 +21,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - [TRAINING] Added RMSPropOptimizer, AdagradOptimizer
 - [NEW OP] shape_n, sparse_softmax_cross_entropy_with_logits, split, unstack
 - Added RNN sample
-  
+
 ### Fixes
 - Fixed gradient computation when passing an array of tensors to a function
 - Added gradients for various other ops

data/lib/tensor_stream.rb

@@ -33,6 +33,7 @@ require 'tensor_stream/utils'
 require 'tensor_stream/train/utils'
 require 'tensor_stream/images'
 require 'tensor_stream/trainer'
+require 'tensor_stream/profile/report_tool'
 
 # require 'tensor_stream/libraries/layers'
 require 'tensor_stream/monkey_patches/integer'
@@ -44,4 +45,8 @@ module TensorStream
   extend TensorStream::Ops
   extend TensorStream::Debugging
   extend TensorStream::Utils
+
+  def self.__version__
+    TensorStream::VERSION
+  end
 end

data/lib/tensor_stream/evaluator/base_evaluator.rb

@@ -19,7 +19,10 @@ module TensorStream
       end
     end
 
+    ##
     # Evaluator base class
+    #
+    # Base class to be used by all tensor_stream evaluators, provides support functions
     class BaseEvaluator
       def initialize(session, _device, thread_pool: nil, log_intermediates: false)
         @session = session
@@ -57,6 +60,7 @@ module TensorStream
         substrs.each do |q|
           components = q.split(':')
           next if components.size.zero?
+
           if components[0] == 'device' # use tensorflow convention
             device_type = components[1]
             select_index = components[2].to_i
@@ -79,6 +83,7 @@ module TensorStream
             evaluator_class = TensorStream::Evaluator.evaluators[components[1]][:class]
             return nil unless self == evaluator_class
             return evaluator_class.fetch_device(components[2..components.size]) if evaluator_class.respond_to?(:fetch_device)
+
             return nil
           end
         end
@@ -121,14 +126,34 @@ module TensorStream
           global_eval(tensor, i, execution_context, op_options)
         end
 
-        instance_exec(execution_context, tensor, resolved_inputs, &op[:block])
+        start_time = if profile_enabled?
+                       time = Time.now
+                       time.to_i * (10**9) + time.nsec
+                     end
+
+        instance_exec(execution_context, tensor, resolved_inputs, &op[:block]).tap do
+          if profile_enabled?
+            time = Time.now
+            end_time = time.to_i * (10**9) + time.nsec
+            @context[:profile] ||= { step: 0, operations: {} }
+            @context[:profile][:step] += 1
+            @context[:profile][:operations][tensor.name] = { op: tensor.operation,
+                                                             step: @context[:profile][:step],
+                                                             eval_time: end_time - start_time,
+                                                             tensor: tensor }
+          end
+        end
       end
 
       protected
 
+      def profile_enabled?
+        @context[:_options][:profile_enabled]
+      end
+
       ##
       # called when passing control to another evaluator
-      def perform_transition(tensor, input, _next_evaluator)
+      def perform_transition(tensor, input, _next_evaluator, execution_context)
         cache_key = "#{tensor.graph.object_id}_#{input.name}:#{object_id}"
         return @context[:_cache][cache_key] if @context[:_cache].key?(cache_key)
 
@@ -141,14 +166,20 @@ module TensorStream
       def global_eval(tensor, input, execution_context, op_options = {})
         return nil unless input
         return input unless input.is_a?(Tensor)
+        # puts "global eval #{tensor.name}"
         @context[:_cache][:placement][input.name] = @session.assign_evaluator(input) if @context[:_cache][:placement][input.name].nil?
-        if object_id != @context[:_cache][:placement][input.name][1].object_id # tensor is on another device or evaluator
-          perform_transition(tensor, input, @context[:_cache][:placement][input.name][1])
+        if !on_same_device?(input) # tensor is on another device or evaluator
+          # puts "transition #{object_id} -> #{@context[:_cache][:placement][input.name][1].object_id}"
+          perform_transition(tensor, input, @context[:_cache][:placement][input.name][1], execution_context)
         else
           prepare_input(input, execution_context, op_options)
         end
       end
 
+      def on_same_device?(tensor)
+        object_id == @context[:_cache][:placement][tensor.name][1].object_id
+      end
+
       def get_broadcast_gradient_args(input_a, input_b)
         return [[], []] if input_a == input_b
 

data/lib/tensor_stream/evaluator/ruby/math_ops.rb

@@ -186,7 +186,6 @@ module TensorStream
             end
             reduced_val
           end
-
           reduction(context, tensor, func)
         end
 

data/lib/tensor_stream/evaluator/ruby_evaluator.rb

@@ -282,15 +282,17 @@ module TensorStream
 
       def eval_operation(tensor, child_context)
         return @context[tensor.name] if @context.key?(tensor.name)
-        invoke(tensor, child_context).tap do |result|
 
+        # puts "ruby eval #{object_id}: #{tensor.name}"
+        invoke(tensor, child_context).tap do |result|
+          # puts "result done ruby #{object_id}: #{tensor.name}"
           # assertions to make sure inferred shapes == actual evaluated shapes
           if tensor.shape.known? && (result.is_a?(Array) || result.is_a?(Float) || result.is_a?(Integer))
             if shape_eval(result) != tensor.shape.shape
               raise "assert error #{tensor.name} #{shape_eval(result)} != #{tensor.shape.shape}"
             end
           end
-  
+
           if tensor.breakpoint
             a = resolve_placeholder(tensor.inputs[0], child_context) if tensor.inputs && tensor.inputs[0]
             b = resolve_placeholder(tensor.inputs[1], child_context) if tensor.inputs && tensor.inputs[1]

data/lib/tensor_stream/profile/report_tool.rb

@@ -0,0 +1,18 @@
+module TensorStream
+  ##
+  # Utiliity functions for creating performance reports
+  class ReportTool
+    def self.profile_for(session, order_by: :slowest)
+      context = session.last_session_context
+      eval_times = context[:profile][:operations].map do |name, profile|
+        [name, profile[:eval_time], profile[:tensor].source]
+      end
+
+      if order_by == :slowest
+        eval_times.sort_by { |a, b| b[1] <=> a[1] }
+      else
+        eval_times.sort_by { |a, b| a[1] <=> b[1] }
+      end
+    end
+  end
+end

data/lib/tensor_stream/session.rb

@@ -6,13 +6,13 @@ module TensorStream
     attr_reader :last_session_context, :closed, :target, :session_cache
     attr_accessor :randomizer
 
-    def initialize(evaluator = nil, thread_pool_class: Concurrent::ImmediateExecutor, log_device_placement: false, evaluator_options: {})
+    def initialize(evaluator = nil, thread_pool_class: Concurrent::ImmediateExecutor, log_device_placement: false, profile_enabled: false, evaluator_options: {})
       @thread_pool = thread_pool_class.new
       @closed = false
       @session_cache = {}
       @randomizer = {}
       @log_device_placement = log_device_placement
-      @evaluator_options = evaluator_options
+      @evaluator_options = evaluator_options.merge(profile_enabled: profile_enabled)
       get_evaluator_classes(evaluator)
       @evaluators = {}
     end
@@ -45,8 +45,13 @@ module TensorStream
                 else
                   {}
                 end
+
+      @evaluator_options[:thread_pool] = @thread_pool
+      @evaluator_options[:log_intermediates] = options[:log_intermediates]
+
       context = {
-        _cache: @session_cache
+        _cache: @session_cache,
+        _options: options.merge(@evaluator_options)
       }
 
       # scan for placeholders and assign value
@@ -56,8 +61,6 @@ module TensorStream
         end
       end
 
-      @evaluator_options[:thread_pool] = @thread_pool
-      @evaluator_options[:log_intermediates] = options[:log_intermediates]
 
       args.each { |t| prepare_evaluators(t, context) }
       @last_session_context = context
@@ -69,6 +72,7 @@ module TensorStream
       end
       result = args.collect do |e|
         next e.value if e.is_a?(Tensor) && e.is_const && e.value
+
         value = delegate_to_evaluator(e, context, {})
         recursive_eval(value)
       end

data/lib/tensor_stream/utils.rb

@@ -129,8 +129,18 @@ module TensorStream
       Thread.current[:tensor_stream_variable_scope].map(&:name).compact.reject(&:empty?).join('/')
     end
 
-    def session(evaluator = nil, thread_pool_class: Concurrent::ImmediateExecutor, log_device_placement: false)
-      session = TensorStream::Session.new(evaluator, thread_pool_class: thread_pool_class, log_device_placement: log_device_placement)
+    ##
+    # Creates a session context where operations can be executed
+    #
+    # Args:
+    #     evaluator: Specific evaluator to use, otherwise the best evaluator will automatically be determined
+    #
+    # Options:
+    #     thread_pool_class: Class to use to manage thread pooling
+    #     log_device_placement: Show assigned device/evalutor for each tensor op
+    #     profile_enabled: Log performance metrics for each operation
+    def session(evaluator = nil, thread_pool_class: Concurrent::ImmediateExecutor, log_device_placement: false, profile_enabled: false)
+      session = TensorStream::Session.new(evaluator, thread_pool_class: thread_pool_class, log_device_placement: log_device_placement, profile_enabled: profile_enabled)
       yield session if block_given?
 
       session
@@ -151,6 +161,7 @@ module TensorStream
 
     def constant(value, dtype: nil, shape: nil, internal: false, name: 'Const')
       shared_options = { const: true, value: value, name: name, internal: internal }
+
       if value.is_a?(Float)
         TensorStream::Tensor.new(dtype || :float32, 0, shape || [], shared_options)
       elsif value.is_a?(Integer)
@@ -162,6 +173,7 @@ module TensorStream
       elsif value.is_a?(Array)
         dimension = shape || shape_eval(value)
         rank = dimension.size
+        TensorStream.check_if_dense(value)
 
         cur_dtype = dtype || Tensor.detect_type(value.flatten.last)
         value = Tensor.cast_dtype(value, cur_dtype) unless dtype.nil?
@@ -224,18 +236,39 @@ module TensorStream
       TensorStream.get_default_graph.control_dependencies(control_inputs, &block)
     end
 
-    def convert_to_tensor(value, dtype: nil, name: nil, preferred_dtype: nil)
+    def convert_to_tensor(value, dtype: nil, name: nil)
       return value if value.is_a?(Tensor)
       return convert_to_tensor(value.call) if value.is_a?(Proc)
+
       if value.is_a?(Array) && value[0].is_a?(Tensor)
         return TensorStream.stack(value) if value.size > 1
 
         return TensorStream.expand_dims(value[0], 0)
       end
 
+      check_if_dense(value)
       i_cons(value, dtype: dtype || Tensor.detect_type(value), name: name)
     end
 
+    ##
+    # Check to make sure passed array is dense
+    #
+    def check_if_dense(value, expected_shape = nil)
+      return unless value.is_a?(Array)
+      return if value.empty?
+
+      expected_shape ||= shape_eval(value)
+
+      s = expected_shape.shift
+      raise TensorStream::ValueError, "Argument must be a dense tensor: #{value}, expected size #{s} got #{value.size}" if value.size != s
+
+      return if expected_shape.empty?
+
+      value.each do |item|
+        check_if_dense(item, expected_shape.dup)
+      end
+    end
+
     def check_allowed_types(input, types)
       return input unless input.is_a?(Tensor)
       return input if input.data_type.nil?

data/lib/tensor_stream/version.rb

@@ -1,5 +1,5 @@
 module TensorStream
-  VERSION = '0.9.1'.freeze
+  VERSION = '0.9.2'.freeze
 
   def self.version
     VERSION

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: tensor_stream
 version: !ruby/object:Gem::Version
-  version: 0.9.1
+  version: 0.9.2
 platform: ruby
 authors:
 - Joseph Emmanuel Dayo
 autorequire: 
 bindir: exe
 cert_chain: []
-date: 2018-10-19 00:00:00.000000000 Z
+date: 2018-10-21 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: bundler
@@ -285,6 +285,7 @@ files:
 - lib/tensor_stream/operation.rb
 - lib/tensor_stream/ops.rb
 - lib/tensor_stream/placeholder.rb
+- lib/tensor_stream/profile/report_tool.rb
 - lib/tensor_stream/session.rb
 - lib/tensor_stream/tensor.rb
 - lib/tensor_stream/tensor_shape.rb
@@ -336,7 +337,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
       version: '0'
 requirements: []
 rubyforge_project: 
-rubygems_version: 2.6.10
+rubygems_version: 2.7.7
 signing_key: 
 specification_version: 4
 summary: A Pure ruby tensorflow implementation