diff options
author | Daryl Ng <darylng@google.com> | 2018-07-27 17:49:39 -0700 |
---|---|---|
committer | TensorFlower Gardener <gardener@tensorflow.org> | 2018-07-27 17:53:14 -0700 |
commit | e56f30652248d83ffce0de048d0c59244b4aa9c4 (patch) | |
tree | 0cd04ff46dda9c5fecb0a4b3ecc9c745036ebc86 /tensorflow | |
parent | 444fae503e93152caf9286686c5ca4cfcaf27b54 (diff) |
Changing syntax of optimization_parameters.proto from proto2 to proto3. Since the lower and upper fields of ClippingLimits used non-zero defaults, they had to be converted to use the google.protobuf.FloatValue wrappers.
PiperOrigin-RevId: 206400787
Diffstat (limited to 'tensorflow')
-rw-r--r-- | tensorflow/contrib/tpu/proto/optimization_parameters.proto | 114 |
1 files changed, 58 insertions, 56 deletions
diff --git a/tensorflow/contrib/tpu/proto/optimization_parameters.proto b/tensorflow/contrib/tpu/proto/optimization_parameters.proto index 9150606f5e..2cc17d6d92 100644 --- a/tensorflow/contrib/tpu/proto/optimization_parameters.proto +++ b/tensorflow/contrib/tpu/proto/optimization_parameters.proto @@ -1,10 +1,12 @@ -syntax = "proto2"; +syntax = "proto3"; package tensorflow.tpu; +import "google/protobuf/wrappers.proto"; + message ClippingLimits { - optional float lower = 1 [default = -inf]; - optional float upper = 2 [default = inf]; + google.protobuf.FloatValue lower = 1; // -inf if not set + google.protobuf.FloatValue upper = 2; // +inf if not set } // Get the learning rate from a <yet to be determined> source that can change @@ -21,18 +23,18 @@ message LearningRate { } message AdagradParameters { - optional float initial_accumulator = 1 [default = 0.]; + float initial_accumulator = 1; } message StochasticGradientDescentParameters { } message FtrlParameters { - optional float l1 = 1 [default = 0.]; - optional float l2 = 2 [default = 0.]; - optional float lr_power = 3 [default = 0.]; - optional float initial_accum = 4 [default = 0.]; - optional float initial_linear = 5 [default = 0.]; + float l1 = 1; + float l2 = 2; + float lr_power = 3; + float initial_accum = 4; + float initial_linear = 5; } // The Adam optimizer does not implement hyper-parameter update; use the dynamic @@ -41,84 +43,84 @@ message FtrlParameters { // Here, t is the current timestep. // https://github.com/tensorflow/tensorflow/blob/ab51450c817674c8ff08a7ae4f8ac50cdc4bed8b/tensorflow/python/training/adam.py#L54 message AdamParameters { - optional float beta1 = 3 [default = 0.]; - optional float beta2 = 4 [default = 0.]; - optional float epsilon = 5 [default = 0.]; - optional float initial_m = 6 [default = 0.]; - optional float initial_v = 7 [default = 0.]; + float beta1 = 3; + float beta2 = 4; + float epsilon = 5; + float initial_m = 6; + float initial_v = 7; } message MomentumParameters { - optional float momentum = 1 [default = 0.]; - optional bool use_nesterov = 2 [default = false]; - optional float initial_accum = 3 [default = 0.]; + float momentum = 1; + bool use_nesterov = 2; + float initial_accum = 3; } message RmsPropParameters { - optional float rho = 1 [default = 0.]; - optional float momentum = 2 [default = 0.]; - optional float epsilon = 3 [default = 0.]; - optional float initial_ms = 4 [default = 0.]; - optional float initial_mom = 5 [default = 0.]; + float rho = 1; + float momentum = 2; + float epsilon = 3; + float initial_ms = 4; + float initial_mom = 5; } message CenteredRmsPropParameters { - optional float rho = 1 [default = 0.]; - optional float momentum = 2 [default = 0.]; - optional float epsilon = 3 [default = 0.]; - optional float initial_ms = 4 [default = 0.]; - optional float initial_mom = 5 [default = 0.]; - optional float initial_mg = 6 [default = 0.]; + float rho = 1; + float momentum = 2; + float epsilon = 3; + float initial_ms = 4; + float initial_mom = 5; + float initial_mg = 6; } message MdlAdagradLightParameters { - optional float l2 = 1; - optional float lr_power = 2; - optional float min_servable_mdl_benefit = 3; - optional float mdl_mix_in_margin = 4; - optional float mdl_benefit_rampup_coeff = 5; - optional float mdl_min_weight = 6; - optional float benefit_revisit_scale = 7; - optional float max_event_benefit = 8; - optional float max_total_benefit = 9; - optional float mdl_hard_limit = 10; - optional bool hard_limit_min_benefit = 11; - optional bool mdl_regularize = 12; - optional float initial_accumulator = 13; - optional float initial_weight = 14; - optional float initial_benefit = 15; + float l2 = 1; + float lr_power = 2; + float min_servable_mdl_benefit = 3; + float mdl_mix_in_margin = 4; + float mdl_benefit_rampup_coeff = 5; + float mdl_min_weight = 6; + float benefit_revisit_scale = 7; + float max_event_benefit = 8; + float max_total_benefit = 9; + float mdl_hard_limit = 10; + bool hard_limit_min_benefit = 11; + bool mdl_regularize = 12; + float initial_accumulator = 13; + float initial_weight = 14; + float initial_benefit = 15; } message AdadeltaParameters { - optional float rho = 1; - optional float epsilon = 2; - optional float initial_accumulator = 3 [default = 0.]; - optional float initial_update = 4 [default = 0.]; + float rho = 1; + float epsilon = 2; + float initial_accumulator = 3; + float initial_update = 4; } message ProximalAdagradParameters { - optional float l1 = 1; - optional float l2 = 2; - optional float initial_accumulator = 3; + float l1 = 1; + float l2 = 2; + float initial_accumulator = 3; } message OptimizationParameters { // Learning rate used for updating the embedding layer parameters. - optional LearningRate learning_rate = 13; + LearningRate learning_rate = 13; reserved 1; // Old learning rate tag. // Limits to which to clip the weight values after the backward pass; not // present means no limits are applied. - optional ClippingLimits clipping_limits = 2; + ClippingLimits clipping_limits = 2; // Limits to which to clip the backward pass gradient before using it for // updates; not present means no limits are applied. - optional ClippingLimits gradient_clipping_limits = 7; + ClippingLimits gradient_clipping_limits = 7; // Whether to use gradient accumulation (do two passes over the input // gradients: one to accumulate them into a temporary array and another to // apply them using the actual optimization algorithm). - optional bool use_gradient_accumulation = 15 [default = false]; + bool use_gradient_accumulation = 15; // Optimization algorithm parameters; which field is selected determines which // algorithm to use. @@ -140,7 +142,7 @@ message OptimizationParameters { // value vector and any extra accumulators, etc.). message StateVariableSpecification { // Parameter name for the state variable. - optional string name = 1; + string name = 1; // A normal state variable that should be saved and restored in checkpoints // and used as an input or output to non-debug TensorFlow ops. @@ -151,7 +153,7 @@ message StateVariableSpecification { // from users (used for intermediate gradients being accumulated, for // example). message FillWithConstant { - optional double initial_value = 1; + double initial_value = 1; } // Usage type of this state variable. |