org.platanios.tensorflow.api.ops.training.distribute.values
Primary variable.
Index map from devices to variables.
Creates an op that assigns the provided value to this variable and returns its value.
Creates an op that assigns the provided value to this variable and returns its value.
Value to assign the variable to.
Name for created op.
Variable value read op, after the assignment.
Creates an op that adds the provided value to the current value of the variable and returns its value.
Creates an op that adds the provided value to the current value of the variable and returns its value.
Value to add to the current variable value.
Name for created op.
Variable value read op, after the addition.
Creates an op that applies updates the provided sparse value updates to this variable and returns its value.
Creates an op that applies updates the provided sparse value updates to this variable and returns its value.
Indices corresponding to the values
used for the update.
Values to use for updating, corresponding to the provided indices
.
Name for created op.
Variable value read op, after the addition.
Creates an op that adds the provided sparse value to the current value of the variable and returns its value.
Creates an op that adds the provided sparse value to the current value of the variable and returns its value.
Indices corresponding to the values
being added.
Values to be added, corresponding to the provided indices
.
Name for created op.
Variable value read op, after the addition.
Creates an op that subtracts the provided sparse value from the current value of the variable and returns its value.
Creates an op that subtracts the provided sparse value from the current value of the variable and returns its value.
Indices corresponding to the values
being subtracted.
Values to be subtracted, corresponding to the provided indices
.
Name for created op.
Variable value read op, after the subtraction.
Creates an op that subtracts the provided value from the current value of the variable and returns its value.
Creates an op that subtracts the provided value from the current value of the variable and returns its value.
Value to subtract from the current variable value.
Name for created op.
Variable value read op, after the subtraction.
Data type of this variable.
Data type of this variable.
Returns the devices on which this value is distributed.
Returns the devices on which this value is distributed.
Type of this distributed value (e.g., per-device or mirrored).
Type of this distributed value (e.g., per-device or mirrored).
Creates an op that reads the value of this variable sparsely, using the provided indices
.
Creates an op that reads the value of this variable sparsely, using the provided indices
.
This method should be used when there are multiple reads, or when it is desirable to read the value only after some condition is true.
Indices to use for the sparse read.
Name for the created op.
Created op.
Returns the value on the specified device (defaults to the current device, if not provided.
Returns the value on the specified device (defaults to the current device, if not provided.
Graph where this variable is defined.
Graph where this variable is defined.
Index map from devices to variables.
Index map from devices to variables.
Value of the initialized variable.
Value of the initialized variable. You should use this instead of the variable itself to initialize another variable with a value that depends on the value of this variable.
Example:
// Initialize `v` with random values, and then use `initializedValue` to guarantee that `v` has been initialized // before its value is used to initialize `w`. The random tensor will only be sampled once. val v = tf.variable("v", FLOAT32, Shape(10, 40), tf.RandomTruncatedNormalInitializer()) val w = tf.variable("w", initializer = tf.ConstantInitializer(v.initializedValue * 2.0))
Returns the initializer for this variable.
Returns the initializer for this variable.
Op output that is true
when the variable has been initialized and false
otherwise.
Op output that is true
when the variable has been initialized and false
otherwise.
Name of this variable.
Name of this variable.
Returns true if the values are distributed on the provided device.
Returns true if the values are distributed on the provided device.
Returns the op of this variable.
Returns the op of this variable.
Primary variable.
Primary variable.
Creates an op that reads the value of this variable.
Creates an op that reads the value of this variable.
This method should be used when there are multiple reads, or when it is desirable to read the value only after some condition is true.
The returned value may be different from that of value depending on the device being used, the control dependencies, etc.
Created op.
Shape of this variable.
Shape of this variable.
Alias for toVariableDef
.
Alias for toVariableDef
.
Converts this object to its corresponding ProtoBuf object.
Converts this object to its corresponding ProtoBuf object.
ProtoBuf object corresponding to this object.
Convert this object to its corresponding ProtoBuf object.
Convert this object to its corresponding ProtoBuf object.
Optional string specifying the name scope to remove. Only the ops within this name scope will be included in the resulting ProtoBuf object and the export scope will be stripped from their names to allow for easy import into new name scopes.
ProtoBuf object corresponding to this object.
Returns a cached op which reads the last value of this partitioned variable.
Returns a cached op which reads the last value of this partitioned variable.
You can not assign a new value to the returned tensor as it is not a reference to the variable.
The returned op output will not inherit the control dependencies from the scope where the value is used, which is equivalent behavior to that of getting the value of a variable.
NOTE: You usually do not need to call this method directly, as all ops that use variables do so by internally converting them to tensors.
Holds a map from devices to synchronized variables.
NOTE: We use
MirroredVariable.getUpdateDevice
for the assignment methods to enforce that we are in anupdate
scope. The arguments toupdate()
are automatically unwrapped and so theupdate()
function would normally see regular variables and not mirrored variables. However, theupdate()
function can still operate on wrapped mirrored variables through object members, captured arguments, etc. This is more likely in anupdateNonSlot()
scope, which can update several non-slot variables in one call.