// Copyright 2017 Google Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

syntax = "proto3";

package google.cloud.ml.v1;

import "google/api/annotations.proto";
import "google/api/auth.proto";
import "google/protobuf/empty.proto";
import "google/protobuf/timestamp.proto";

option go_package = "google.golang.org/genproto/googleapis/cloud/ml/v1;ml";
option java_multiple_files = true;
option java_outer_classname = "JobServiceProto";
option java_package = "com.google.cloud.ml.api.v1";

// Copyright 2017 Google Inc. All Rights Reserved.
//
// Proto file for the Google Cloud Machine Learning Engine.
// Describes the 'job service' to manage training and prediction jobs.


// Service to create and manage training and batch prediction jobs.
service JobService {
  // Creates a training or a batch prediction job.
  rpc CreateJob(CreateJobRequest) returns (Job) {
    option (google.api.http) = { post: "/v1/{parent=projects/*}/jobs" body: "job" };
  }

  // Lists the jobs in the project.
  rpc ListJobs(ListJobsRequest) returns (ListJobsResponse) {
    option (google.api.http) = { get: "/v1/{parent=projects/*}/jobs" };
  }

  // Describes a job.
  rpc GetJob(GetJobRequest) returns (Job) {
    option (google.api.http) = { get: "/v1/{name=projects/*/jobs/*}" };
  }

  // Cancels a running job.
  rpc CancelJob(CancelJobRequest) returns (google.protobuf.Empty) {
    option (google.api.http) = { post: "/v1/{name=projects/*/jobs/*}:cancel" body: "*" };
  }
}

// Represents input parameters for a training job.
message TrainingInput {
  // A scale tier is an abstract representation of the resources Cloud ML
  // will allocate to a training job. When selecting a scale tier for your
  // training job, you should consider the size of your training dataset and
  // the complexity of your model. As the tiers increase, virtual machines are
  // added to handle your job, and the individual machines in the cluster
  // generally have more memory and greater processing power than they do at
  // lower tiers. The number of training units charged per hour of processing
  // increases as tiers get more advanced. Refer to the
  // [pricing guide](/ml/pricing) for more details. Note that in addition to
  // incurring costs, your use of training resources is constrained by the
  // [quota policy](/ml/quota).
  enum ScaleTier {
    // A single worker instance. This tier is suitable for learning how to use
    // Cloud ML, and for experimenting with new models using small datasets.
    BASIC = 0;

    // Many workers and a few parameter servers.
    STANDARD_1 = 1;

    // A large number of workers with many parameter servers.
    PREMIUM_1 = 3;

    // A single worker instance [with a GPU](ml/docs/how-tos/using-gpus).
    BASIC_GPU = 6;

    // The CUSTOM tier is not a set tier, but rather enables you to use your
    // own cluster specification. When you use this tier, set values to
    // configure your processing cluster according to these guidelines:
    //
    // *   You _must_ set `TrainingInput.masterType` to specify the type
    //     of machine to use for your master node. This is the only required
    //     setting.
    //
    // *   You _may_ set `TrainingInput.workerCount` to specify the number of
    //     workers to use. If you specify one or more workers, you _must_ also
    //     set `TrainingInput.workerType` to specify the type of machine to use
    //     for your worker nodes.
    //
    // *   You _may_ set `TrainingInput.parameterServerCount` to specify the
    //     number of parameter servers to use. If you specify one or more
    //     parameter servers, you _must_ also set
    //     `TrainingInput.parameterServerType` to specify the type of machine to
    //     use for your parameter servers.
    //
    // Note that all of your workers must use the same machine type, which can
    // be different from your parameter server type and master type. Your
    // parameter servers must likewise use the same machine type, which can be
    // different from your worker type and master type.
    CUSTOM = 5;
  }

  // Required. Specifies the machine types, the number of replicas for workers
  // and parameter servers.
  ScaleTier scale_tier = 1;

  // Optional. Specifies the type of virtual machine to use for your training
  // job's master worker.
  //
  // The following types are supported:
  //
  // <dl>
  //   <dt>standard</dt>
  //   <dd>
  //   A basic machine configuration suitable for training simple models with
  //   small to moderate datasets.
  //   </dd>
  //   <dt>large_model</dt>
  //   <dd>
  //   A machine with a lot of memory, specially suited for parameter servers
  //   when your model is large (having many hidden layers or layers with very
  //   large numbers of nodes).
  //   </dd>
  //   <dt>complex_model_s</dt>
  //   <dd>
  //   A machine suitable for the master and workers of the cluster when your
  //   model requires more computation than the standard machine can handle
  //   satisfactorily.
  //   </dd>
  //   <dt>complex_model_m</dt>
  //   <dd>
  //   A machine with roughly twice the number of cores and roughly double the
  //   memory of <code suppresswarning="true">complex_model_s</code>.
  //   </dd>
  //   <dt>complex_model_l</dt>
  //   <dd>
  //   A machine with roughly twice the number of cores and roughly double the
  //   memory of <code suppresswarning="true">complex_model_m</code>.
  //   </dd>
  //   <dt>standard_gpu</dt>
  //   <dd>
  //   A machine equivalent to <code suppresswarning="true">standard</code> that
  //   also includes a
  //   <a href="ml/docs/how-tos/using-gpus">
  //   GPU that you can use in your trainer</a>.
  //   </dd>
  //   <dt>complex_model_m_gpu</dt>
  //   <dd>
  //   A machine equivalent to
  //   <code suppresswarning="true">coplex_model_m</code> that also includes
  //   four GPUs.
  //   </dd>
  // </dl>
  //
  // You must set this value when `scaleTier` is set to `CUSTOM`.
  string master_type = 2;

  // Optional. Specifies the type of virtual machine to use for your training
  // job's worker nodes.
  //
  // The supported values are the same as those described in the entry for
  // `masterType`.
  //
  // This value must be present when `scaleTier` is set to `CUSTOM` and
  // `workerCount` is greater than zero.
  string worker_type = 3;

  // Optional. Specifies the type of virtual machine to use for your training
  // job's parameter server.
  //
  // The supported values are the same as those described in the entry for
  // `master_type`.
  //
  // This value must be present when `scaleTier` is set to `CUSTOM` and
  // `parameter_server_count` is greater than zero.
  string parameter_server_type = 4;

  // Optional. The number of worker replicas to use for the training job. Each
  // replica in the cluster will be of the type specified in `worker_type`.
  //
  // This value can only be used when `scale_tier` is set to `CUSTOM`. If you
  // set this value, you must also set `worker_type`.
  int64 worker_count = 5;

  // Optional. The number of parameter server replicas to use for the training
  // job. Each replica in the cluster will be of the type specified in
  // `parameter_server_type`.
  //
  // This value can only be used when `scale_tier` is set to `CUSTOM`.If you
  // set this value, you must also set `parameter_server_type`.
  int64 parameter_server_count = 6;

  // Required. The Google Cloud Storage location of the packages with
  // the training program and any additional dependencies.
  repeated string package_uris = 7;

  // Required. The Python module name to run after installing the packages.
  string python_module = 8;

  // Optional. Command line arguments to pass to the program.
  repeated string args = 10;

  // Optional. The set of Hyperparameters to tune.
  HyperparameterSpec hyperparameters = 12;

  // Required. The Google Compute Engine region to run the training job in.
  string region = 14;

  // Optional. A Google Cloud Storage path in which to store training outputs
  // and other data needed for training. This path is passed to your TensorFlow
  // program as the 'job_dir' command-line argument. The benefit of specifying
  // this field is that Cloud ML validates the path for use in training.
  string job_dir = 16;

  // Optional. The Google Cloud ML runtime version to use for training.  If not
  // set, Google Cloud ML will choose the latest stable version.
  string runtime_version = 15;
}

// Represents a set of hyperparameters to optimize.
message HyperparameterSpec {
  // The available types of optimization goals.
  enum GoalType {
    // Goal Type will default to maximize.
    GOAL_TYPE_UNSPECIFIED = 0;

    // Maximize the goal metric.
    MAXIMIZE = 1;

    // Minimize the goal metric.
    MINIMIZE = 2;
  }

  // Required. The type of goal to use for tuning. Available types are
  // `MAXIMIZE` and `MINIMIZE`.
  //
  // Defaults to `MAXIMIZE`.
  GoalType goal = 1;

  // Required. The set of parameters to tune.
  repeated ParameterSpec params = 2;

  // Optional. How many training trials should be attempted to optimize
  // the specified hyperparameters.
  //
  // Defaults to one.
  int32 max_trials = 3;

  // Optional. The number of training trials to run concurrently.
  // You can reduce the time it takes to perform hyperparameter tuning by adding
  // trials in parallel. However, each trail only benefits from the information
  // gained in completed trials. That means that a trial does not get access to
  // the results of trials running at the same time, which could reduce the
  // quality of the overall optimization.
  //
  // Each trial will use the same scale tier and machine types.
  //
  // Defaults to one.
  int32 max_parallel_trials = 4;

  // Optional. The Tensorflow summary tag name to use for optimizing trials. For
  // current versions of Tensorflow, this tag name should exactly match what is
  // shown in Tensorboard, including all scopes.  For versions of Tensorflow
  // prior to 0.12, this should be only the tag passed to tf.Summary.
  // By default, "training/hptuning/metric" will be used.
  string hyperparameter_metric_tag = 5;
}

// Represents a single hyperparameter to optimize.
message ParameterSpec {
  // The type of the parameter.
  enum ParameterType {
    // You must specify a valid type. Using this unspecified type will result in
    // an error.
    PARAMETER_TYPE_UNSPECIFIED = 0;

    // Type for real-valued parameters.
    DOUBLE = 1;

    // Type for integral parameters.
    INTEGER = 2;

    // The parameter is categorical, with a value chosen from the categories
    // field.
    CATEGORICAL = 3;

    // The parameter is real valued, with a fixed set of feasible points. If
    // `type==DISCRETE`, feasible_points must be provided, and
    // {`min_value`, `max_value`} will be ignored.
    DISCRETE = 4;
  }

  // The type of scaling that should be applied to this parameter.
  enum ScaleType {
    // By default, no scaling is applied.
    NONE = 0;

    // Scales the feasible space to (0, 1) linearly.
    UNIT_LINEAR_SCALE = 1;

    // Scales the feasible space logarithmically to (0, 1). The entire feasible
    // space must be strictly positive.
    UNIT_LOG_SCALE = 2;

    // Scales the feasible space "reverse" logarithmically to (0, 1). The result
    // is that values close to the top of the feasible space are spread out more
    // than points near the bottom. The entire feasible space must be strictly
    // positive.
    UNIT_REVERSE_LOG_SCALE = 3;
  }

  // Required. The parameter name must be unique amongst all ParameterConfigs in
  // a HyperparameterSpec message. E.g., "learning_rate".
  string parameter_name = 1;

  // Required. The type of the parameter.
  ParameterType type = 4;

  // Required if type is `DOUBLE` or `INTEGER`. This field
  // should be unset if type is `CATEGORICAL`. This value should be integers if
  // type is INTEGER.
  double min_value = 2;

  // Required if typeis `DOUBLE` or `INTEGER`. This field
  // should be unset if type is `CATEGORICAL`. This value should be integers if
  // type is `INTEGER`.
  double max_value = 3;

  // Required if type is `CATEGORICAL`. The list of possible categories.
  repeated string categorical_values = 5;

  // Required if type is `DISCRETE`.
  // A list of feasible points.
  // The list should be in strictly increasing order. For instance, this
  // parameter might have possible settings of 1.5, 2.5, and 4.0. This list
  // should not contain more than 1,000 values.
  repeated double discrete_values = 6;

  // Optional. How the parameter should be scaled to the hypercube.
  // Leave unset for categorical parameters.
  // Some kind of scaling is strongly recommended for real or integral
  // parameters (e.g., `UNIT_LINEAR_SCALE`).
  ScaleType scale_type = 7;
}

// Represents the result of a single hyperparameter tuning trial from a
// training job. The TrainingOutput object that is returned on successful
// completion of a training job with hyperparameter tuning includes a list
// of HyperparameterOutput objects, one for each successful trial.
message HyperparameterOutput {
  // An observed value of a metric.
  message HyperparameterMetric {
    // The global training step for this metric.
    int64 training_step = 1;

    // The objective value at this training step.
    double objective_value = 2;
  }

  // The trial id for these results.
  string trial_id = 1;

  // The hyperparameters given to this trial.
  map<string, string> hyperparameters = 2;

  // The final objective metric seen for this trial.
  HyperparameterMetric final_metric = 3;

  // All recorded object metrics for this trial.
  repeated HyperparameterMetric all_metrics = 4;
}

// Represents results of a training job. Output only.
message TrainingOutput {
  // The number of hyperparameter tuning trials that completed successfully.
  // Only set for hyperparameter tuning jobs.
  int64 completed_trial_count = 1;

  // Results for individual Hyperparameter trials.
  // Only set for hyperparameter tuning jobs.
  repeated HyperparameterOutput trials = 2;

  // The amount of ML units consumed by the job.
  double consumed_ml_units = 3;

  // Whether this job is a hyperparameter tuning job.
  bool is_hyperparameter_tuning_job = 4;
}

// Represents input parameters for a prediction job.
message PredictionInput {
  // The format used to separate data instances in the source files.
  enum DataFormat {
    // Unspecified format.
    DATA_FORMAT_UNSPECIFIED = 0;

    // The source file is a text file with instances separated by the
    // new-line character.
    TEXT = 1;

    // The source file is a TFRecord file.
    TF_RECORD = 2;

    // The source file is a GZIP-compressed TFRecord file.
    TF_RECORD_GZIP = 3;
  }

  // Required. The model or the version to use for prediction.
  oneof model_version {
    // Use this field if you want to use the default version for the specified
    // model. The string must use the following format:
    //
    // `"projects/<var>[YOUR_PROJECT]</var>/models/<var>[YOUR_MODEL]</var>"`
    string model_name = 1;

    // Use this field if you want to specify a version of the model to use. The
    // string is formatted the same way as `model_version`, with the addition
    // of the version information:
    //
    // `"projects/<var>[YOUR_PROJECT]</var>/models/<var>YOUR_MODEL/versions/<var>[YOUR_VERSION]</var>"`
    string version_name = 2;

    // Use this field if you want to specify a Google Cloud Storage path for
    // the model to use.
    string uri = 9;
  }

  // Required. The format of the input data files.
  DataFormat data_format = 3;

  // Required. The Google Cloud Storage location of the input data files.
  // May contain wildcards.
  repeated string input_paths = 4;

  // Required. The output Google Cloud Storage location.
  string output_path = 5;

  // Optional. The maximum number of workers to be used for parallel processing.
  // Defaults to 10 if not specified.
  int64 max_worker_count = 6;

  // Required. The Google Compute Engine region to run the prediction job in.
  string region = 7;

  // Optional. The Google Cloud ML runtime version to use for this batch
  // prediction. If not set, Google Cloud ML will pick the runtime version used
  // during the CreateVersion request for this model version, or choose the
  // latest stable version when model version information is not available
  // such as when the model is specified by uri.
  string runtime_version = 8;
}

// Represents results of a prediction job.
message PredictionOutput {
  // The output Google Cloud Storage location provided at the job creation time.
  string output_path = 1;

  // The number of generated predictions.
  int64 prediction_count = 2;

  // The number of data instances which resulted in errors.
  int64 error_count = 3;

  // Node hours used by the batch prediction job.
  double node_hours = 4;
}

// Represents a training or prediction job.
message Job {
  // Describes the job state.
  enum State {
    // The job state is unspecified.
    STATE_UNSPECIFIED = 0;

    // The job has been just created and processing has not yet begun.
    QUEUED = 1;

    // The service is preparing to run the job.
    PREPARING = 2;

    // The job is in progress.
    RUNNING = 3;

    // The job completed successfully.
    SUCCEEDED = 4;

    // The job failed.
    // `error_message` should contain the details of the failure.
    FAILED = 5;

    // The job is being cancelled.
    // `error_message` should describe the reason for the cancellation.
    CANCELLING = 6;

    // The job has been cancelled.
    // `error_message` should describe the reason for the cancellation.
    CANCELLED = 7;
  }

  // Required. The user-specified id of the job.
  string job_id = 1;

  // Required. Parameters to create a job.
  oneof input {
    // Input parameters to create a training job.
    TrainingInput training_input = 2;

    // Input parameters to create a prediction job.
    PredictionInput prediction_input = 3;
  }

  // Output only. When the job was created.
  google.protobuf.Timestamp create_time = 4;

  // Output only. When the job processing was started.
  google.protobuf.Timestamp start_time = 5;

  // Output only. When the job processing was completed.
  google.protobuf.Timestamp end_time = 6;

  // Output only. The detailed state of a job.
  State state = 7;

  // Output only. The details of a failure or a cancellation.
  string error_message = 8;

  // Output only. The current result of the job.
  oneof output {
    // The current training job result.
    TrainingOutput training_output = 9;

    // The current prediction job result.
    PredictionOutput prediction_output = 10;
  }
}

// Request message for the CreateJob method.
message CreateJobRequest {
  // Required. The project name.
  //
  // Authorization: requires `Editor` role on the specified project.
  string parent = 1;

  // Required. The job to create.
  Job job = 2;
}

// Request message for the ListJobs method.
message ListJobsRequest {
  // Required. The name of the project for which to list jobs.
  //
  // Authorization: requires `Viewer` role on the specified project.
  string parent = 1;

  // Optional. Specifies the subset of jobs to retrieve.
  string filter = 2;

  // Optional. A page token to request the next page of results.
  //
  // You get the token from the `next_page_token` field of the response from
  // the previous call.
  string page_token = 4;

  // Optional. The number of jobs to retrieve per "page" of results. If there
  // are more remaining results than this number, the response message will
  // contain a valid value in the `next_page_token` field.
  //
  // The default value is 20, and the maximum page size is 100.
  int32 page_size = 5;
}

// Response message for the ListJobs method.
message ListJobsResponse {
  // The list of jobs.
  repeated Job jobs = 1;

  // Optional. Pass this token as the `page_token` field of the request for a
  // subsequent call.
  string next_page_token = 2;
}

// Request message for the GetJob method.
message GetJobRequest {
  // Required. The name of the job to get the description of.
  //
  // Authorization: requires `Viewer` role on the parent project.
  string name = 1;
}

// Request message for the CancelJob method.
message CancelJobRequest {
  // Required. The name of the job to cancel.
  //
  // Authorization: requires `Editor` role on the parent project.
  string name = 1;
}