d3lm

class Placeholder():
  def __init__(self):
    self.value = None
    _default_graph.placeholders.append(self)

d3lm

class Variable():
  def __init__(self, initial_value=None):
    self.value = initial_value
    _default_graph.variables.append(self)

d3lm

def topology_sort(operation):
    ordering = []
    visited_nodes = set()

    def recursive_helper(node):
      if isinstance(node, Operation):
        for input_node in node.input_nodes:
          if input_node not in visited_nodes:
            recursive_helper(input_node)

d3lm

class Session():
  def run(self, operation, feed_dict={}):
    nodes_sorted = topology_sort(operation)

    for node in nodes_sorted:
      if type(node) == Placeholder:
        node.output = feed_dict[node]
      elif type(node) == Variable or type(node) == Constant:
        node.output = node.value
      else:

d3lm

import tf_api as tf

# create default graph
tf.Graph().as_default()

# construct computational graph by creating some nodes
a = tf.Constant(15)
b = tf.Constant(5)
prod = tf.multiply(a, b)
sum = tf.add(a, b)

d3lm

class BinaryOperation(Operation):
  def __init__(self, a, b):
    super().__init__([a, b])

d3lm

class multiply(BinaryOperation):
  """
  Computes a * b, element-wise
  """
  def forward(self, a, b):
    return a * b

  def backward(self, upstream_grad):
    raise NotImplementedError

d3lm

import numpy as np
import tf_api as tf

# Create model
model = lambda x: 1.23 * x + 10

# Create small dataset of 5 entries
train_X = np.linspace(0, 5, 5).reshape((-1, 1))
train_Y = np.array([model(x) for x in train_X]).reshape((-1, 1))

d3lm

./configure --ninja --node-builtin-modules-path `pwd`

d3lm

CLang optimizations on Mac OSX

Version:

Apple LLVM version 6.0 (clang-600.0.57) (based on LLVM 3.5svn)
Target: x86_64-apple-darwin14.1.0
Thread model: posix

This was made with commands: