def recursion():
print("Inside recursion function")
tf.reset_default_graph()
g = tf.Graph()
firstG = g
with g.as_default():
def body(x):
a = tf.random_uniform(shape=[2, 2], dtype=tf.int32, maxval=100)
b = tf.constant(np.array([[1, 2], [3, 4]]), dtype=tf.int32)
c = a + b
return tf.nn.relu(x + c)
def condition(x):
return tf.reduce_sum(x) < 100
@function.Defun(*[tf.int32])
def fun(arg):
x = tf.Variable(tf.constant(arg, shape=[2, 2]))
return tf.constant(2)
# tf.while_loop(condition,body,[x])
arg1 = tf.constant([1, 1], [3, 3])
fun2run = gen_functional_ops.StatefulPartitionedCall(args=[arg1],
Tout=[tf.int32],
f=fun)
with tf.Session(
graph=g
) as sess: #If you do not pass the graph here, the session would not know about 'g' graph operations/funcs
print("sess is in scope of 'g' graph ") if (
g is firstG) else print("sess is not in scope of 'g' graph")
sess.run(tf.global_variables_initializer())
out = sess.run(fun2run)
print("recursion output is = ", out)
def algrebraicExp():
print("Inside algrebraicExp function")
tf.reset_default_graph()
g = tf.Graph()
firstG = g
with g.as_default():
@function.Defun(*[tf.int32] * 2, func_name="fun1Fn")
def fun1(x, y):
temp = tf.add(x, x)
z = tf.multiply(temp, y)
return z
arg1 = tf.constant(10)
arg2 = tf.constant([20])
fun2run = gen_functional_ops.StatefulPartitionedCall(args=[arg1, arg2],
Tout=[tf.int32],
f=fun1)
with tf.Session(
graph=g
) as sess: #If you do not pass the graph here, the session would not know about 'g' graph operations/funcs
print("sess is in scope of 'g' graph ") if (
g is firstG) else print("sess is not in scope of 'g' graph")
sess.run(tf.global_variables_initializer())
out = sess.run(fun2run)
print("Algebraic invocation output is = ", out)
def functionInvocation():
print("Inside functionInvocation function")
tf.reset_default_graph()
g = tf.Graph()
firstG = g
with g.as_default():
def fun1(x, y):
return tf.multiply(tf.constant(10), (x + y))
@function.Defun(tf.int32, tf.int32, func_name="fun2Fn")
def fun2(x, y):
return fun1(x, y)
fun2run = gen_functional_ops.StatefulPartitionedCall(
args=[tf.constant(10), tf.constant(20)], Tout=[tf.int32], f=fun2)
with tf.Session(
graph=g
) as sess: #If you do not pass the graph here, the session would not know about 'g' graph operations/funcs
print("sess is in scope of 'g' graph ") if (
g is firstG) else print("sess is not in scope of 'g' graph")
sess.run(tf.global_variables_initializer())
out = sess.run(fun2run)
print("Function Invocation output is = ", out)
def doesGraphContainResourceVariables():
mylist = None
tf.reset_default_graph()
g = ops.Graph()
with g.as_default():
@function.Defun(tf.int32)
def Forward(x):
a = constant_op.constant(2)
c = tf.multiply(x, a)
return c
x = tf.constant(35, name='x')
resVar = tf.Variable(x + 5, use_resource=True, name='res')
z = gen_functional_ops.StatefulPartitionedCall(args=[resVar],
Tout=[tf.int32],
f=Forward)
with tf.Session(graph=g) as sess:
sess.run(tf.global_variables_initializer())
sess.run(z)
mylist = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
resVarFound = False
for var in mylist:
if isResourceVariable(var):
print("Found a resource variable")
resVarFound = True
# break
else:
continue
print(resVarFound)
for var in mylist:
print(var, var.__class__)
resVarCnt = len(mylist)
print("Graph contains", resVarCnt, "number of resource variables")
def constants():
print("Inside constants function")
tf.reset_default_graph()
g = tf.Graph()
firstG = g
with g.as_default():
@function.Defun(*[tf.int32] * 2, func_name="fun1Fn")
def fun1(x, y):
return tf.multiply(x, y)
# fun2run = fun1()
arg1 = tf.constant([[1, 2], [3, 4]])
arg2 = tf.constant([[2, 2], [4, 5]])
fun2run = gen_functional_ops.StatefulPartitionedCall(args=[arg1, arg2],
Tout=[tf.int32],
f=fun1)
with tf.Session(
graph=g
) as sess: #If you do not pass the graph here, the session would not know about 'g' graph operations/funcs
print("sess is in scope of 'g' graph ") if (
g is firstG) else print("sess is not in scope of 'g' graph")
sess.run(tf.global_variables_initializer())
out = sess.run(fun2run)
print("constants invocation output is = ", out)
def isDeviceAssignmentConsistent():
print("Inside device assignment function")
tf.reset_default_graph()
g = tf.Graph()
with g.as_default():
@function.Defun()
def Forward():
with ops.device("/CPU:0"):
stfn1 = tf.random_uniform((2, 2), 0.0, 5.0, name="fn1")
with ops.device("/job:localhost/replica:0/task:0/device:CPU:0"):
stfn2 = tf.ones((2, 2), name="fn3")
with ops.device("/job:localhost/replica:0/task:0/device:CPU:0"):
return stfn1 + stfn2
z = gen_functional_ops.StatefulPartitionedCall(args=[],
Tout=[tf.float32],
f=Forward)
run_options = config_pb2.RunOptions(
trace_level=config_pb2.RunOptions.FULL_TRACE)
run_metadata = config_pb2.RunMetadata()
with tf.Session(
graph=g,
config=config_pb2.ConfigProto(device_count={"CPU": 1})) as sess:
sess.run(tf.global_variables_initializer())
print("The output of device assignment run is = ",
sess.run(z, options=run_options, run_metadata=run_metadata))
assignedDevicesSet = set()
for func in run_metadata.step_stats.dev_stats:
print("device used: ", repr(func.device))
assignedDevicesSet.add(func.device)
print("Device assignment inconsistent"
) if len(assignedDevicesSet) > 2 else print(
"Device assignment is consistent")
def placeholder():
print("Inside placeholder function")
tf.reset_default_graph()
g = tf.Graph()
with g.as_default():
@function.Defun(*[tf.int32] * 2)
def Forward(x, y):
#Do not create placeholders in Defun methods..placeholders should be created outside of Defun()..and can be passed inside it
print(x.name)
print(y.name)
b = tf.add(x, y)
return b
pl1 = tf.placeholder(tf.int32, name="pl1")
pl2 = tf.placeholder(tf.int32, name="pl2")
data = np.array([[-1, 1], [2, -2]], dtype=np.int32)
data2 = np.array([[-2, 3], [4, -6]], dtype=np.int32)
z = gen_functional_ops.StatefulPartitionedCall(args=[pl1, pl2],
Tout=[tf.int32],
f=Forward)
feed = {"pl1:0": data, "pl2:0": data2}
with tf.Session(graph=g) as sess:
sess.run(tf.global_variables_initializer())
print("The output of placeholder run is = ", sess.run(z, feed))
def controlFlow():
print("Inside controlFlow function")
tf.reset_default_graph()
g = tf.Graph()
firstG = g
with g.as_default():
@function.Defun(tf.int32, tf.int32)
def fun2(x, y):
val = 10
for i in range(20):
i += 1
if i % 8 == 0:
retVal = tf.constant(i)
break
return retVal
arg1 = tf.constant(7)
arg2 = tf.constant(15)
fun2run = gen_functional_ops.StatefulPartitionedCall(args=[7, 15],
Tout=[tf.int32],
f=fun2)
with tf.Session(
graph=g
) as sess: #If you do not pass the graph here, the session would not know about 'g' graph operations/funcs
print("sess is in scope of 'g' graph ") if (
g is firstG) else print("sess is not in scope of 'g' graph")
sess.run(tf.global_variables_initializer())
out = sess.run(fun2run)
print("control flow output is = ", out)
def variables():
print("Inside variables function")
tf.reset_default_graph()
g = tf.Graph()
with g.as_default():
@function.Defun(tf.int32, tf.int32)
def Forward(x, y):
#create variables outside Defun() method, you can pass variables inside Defun method though
return tf.multiply(x, y)
const1 = tf.constant(10)
const2 = tf.constant(20)
var1 = tf.Variable(const1, dtype=tf.int32)
var2 = tf.Variable(const2, dtype=tf.int32)
z = gen_functional_ops.StatefulPartitionedCall(args=[var1, var2],
Tout=[tf.int32],
f=Forward)
with tf.Session(graph=g) as sess:
sess.run(tf.global_variables_initializer())
print("The output of variables run is = ", sess.run(z))
def doesGraphContainsStatefulOps():
print("Inside StatefulOps function")
tf.reset_default_graph()
g = tf.Graph()
with g.as_default():
@function.Defun()
def Forward():
stfn1 = tf.random_uniform((2, 2), 0.0, 5.0, name="fn1")
stfn2 = tf.random_uniform((2, 2), 0.0, 5.0, name="fn1")
stfn3 = tf.ones((2, 2), name="fn3")
return stfn1 + stfn2 + stfn3
z = gen_functional_ops.StatefulPartitionedCall(args=[],
Tout=[tf.float32],
f=Forward)
with tf.Session(graph=g) as sess:
sess.run(tf.global_variables_initializer())
print("The output of stateful ops run is = ", sess.run(z))
stateful_ops = [(op.name, op.type)
for op in sess.graph.get_operations()
if op.op_def.is_stateful]
print("Graph contains {0} stateful_ops: ", len(stateful_ops))