diff --git a/com.unity.ml-agents/CHANGELOG.md b/com.unity.ml-agents/CHANGELOG.md index ab6ef40cca..fb08f6c330 100755 --- a/com.unity.ml-agents/CHANGELOG.md +++ b/com.unity.ml-agents/CHANGELOG.md @@ -11,6 +11,8 @@ and this project adheres to ### Major Changes #### com.unity.ml-agents (C#) #### ml-agents / ml-agents-envs / gym-unity (Python) +- The Parameter Randomization feature has been refactored to enable sampling of new parameters per episode to improve robustness. The + `resampling-interval` parameter has been removed and the config structure updated. More information [here](https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Training-ML-Agents.md). (#4065) ### Minor Changes #### com.unity.ml-agents (C#) diff --git a/com.unity.ml-agents/Runtime/Sampler.cs b/com.unity.ml-agents/Runtime/Sampler.cs new file mode 100644 index 0000000000..fc48f3c271 --- /dev/null +++ b/com.unity.ml-agents/Runtime/Sampler.cs @@ -0,0 +1,70 @@ +using System; +using System.Collections.Generic; +using Unity.MLAgents.Inference.Utils; +using UnityEngine; +using Random=System.Random; + +namespace Unity.MLAgents +{ + + /// + /// Takes a list of floats that encode a sampling distribution and returns the sampling function. + /// + internal static class SamplerFactory + { + + public static Func CreateUniformSampler(float min, float max, int seed) + { + Random distr = new Random(seed); + return () => min + (float)distr.NextDouble() * (max - min); + } + + public static Func CreateGaussianSampler(float mean, float stddev, int seed) + { + RandomNormal distr = new RandomNormal(seed, mean, stddev); + return () => (float)distr.NextDouble(); + } + + public static Func CreateMultiRangeUniformSampler(IList intervals, int seed) + { + //RNG + Random distr = new Random(seed); + // Will be used to normalize intervalFuncs + float sumIntervalSizes = 0; + //The number of intervals + int numIntervals = (int)(intervals.Count/2); + // List that will store interval lengths + float[] intervalSizes = new float[numIntervals]; + // List that will store uniform distributions + IList> intervalFuncs = new Func[numIntervals]; + // Collect all intervals and store as uniform distrus + // Collect all interval sizes + for(int i = 0; i < numIntervals; i++) + { + var min = intervals[2 * i]; + var max = intervals[2 * i + 1]; + var intervalSize = max - min; + sumIntervalSizes += intervalSize; + intervalSizes[i] = intervalSize; + intervalFuncs[i] = () => min + (float)distr.NextDouble() * intervalSize; + } + // Normalize interval lengths + for(int i = 0; i < numIntervals; i++) + { + intervalSizes[i] = intervalSizes[i] / sumIntervalSizes; + } + // Build cmf for intervals + for(int i = 1; i < numIntervals; i++) + { + intervalSizes[i] += intervalSizes[i - 1]; + } + Multinomial intervalDistr = new Multinomial(seed + 1); + float MultiRange() + { + int sampledInterval = intervalDistr.Sample(intervalSizes); + return intervalFuncs[sampledInterval].Invoke(); + } + return MultiRange; + } + } +} diff --git a/com.unity.ml-agents/Runtime/Sampler.cs.meta b/com.unity.ml-agents/Runtime/Sampler.cs.meta new file mode 100644 index 0000000000..950e28c5b6 --- /dev/null +++ b/com.unity.ml-agents/Runtime/Sampler.cs.meta @@ -0,0 +1,11 @@ +fileFormatVersion: 2 +guid: 39ce0ea5a8b2e47f696f6efc807029f6 +MonoImporter: + externalObjects: {} + serializedVersion: 2 + defaultReferences: [] + executionOrder: 0 + icon: {instanceID: 0} + userData: + assetBundleName: + assetBundleVariant: diff --git a/com.unity.ml-agents/Runtime/SideChannels/EnvironmentParametersChannel.cs b/com.unity.ml-agents/Runtime/SideChannels/EnvironmentParametersChannel.cs index 5c9fd509b6..d28f84d2da 100644 --- a/com.unity.ml-agents/Runtime/SideChannels/EnvironmentParametersChannel.cs +++ b/com.unity.ml-agents/Runtime/SideChannels/EnvironmentParametersChannel.cs @@ -9,7 +9,30 @@ namespace Unity.MLAgents.SideChannels /// internal enum EnvironmentDataTypes { - Float = 0 + Float = 0, + Sampler = 1 + } + + /// + /// The types of distributions from which to sample reset parameters. + /// + internal enum SamplerType + { + /// + /// Samples a reset parameter from a uniform distribution. + /// + Uniform = 0, + + /// + /// Samples a reset parameter from a Gaussian distribution. + /// + Gaussian = 1, + + /// + /// Samples a reset parameter from a MultiRangeUniform distribution. + /// + MultiRangeUniform = 2 + } /// @@ -18,7 +41,7 @@ internal enum EnvironmentDataTypes /// internal class EnvironmentParametersChannel : SideChannel { - Dictionary m_Parameters = new Dictionary(); + Dictionary> m_Parameters = new Dictionary>(); Dictionary> m_RegisteredActions = new Dictionary>(); @@ -42,12 +65,40 @@ protected override void OnMessageReceived(IncomingMessage msg) { var value = msg.ReadFloat32(); - m_Parameters[key] = value; + m_Parameters[key] = () => value; Action action; m_RegisteredActions.TryGetValue(key, out action); action?.Invoke(value); } + else if ((int)EnvironmentDataTypes.Sampler == type) + { + int seed = msg.ReadInt32(); + int samplerType = msg.ReadInt32(); + Func sampler = () => 0.0f; + if ((int)SamplerType.Uniform == samplerType) + { + float min = msg.ReadFloat32(); + float max = msg.ReadFloat32(); + sampler = SamplerFactory.CreateUniformSampler(min, max, seed); + } + else if ((int)SamplerType.Gaussian == samplerType) + { + float mean = msg.ReadFloat32(); + float stddev = msg.ReadFloat32(); + + sampler = SamplerFactory.CreateGaussianSampler(mean, stddev, seed); + } + else if ((int)SamplerType.MultiRangeUniform == samplerType) + { + IList intervals = msg.ReadFloatList(); + sampler = SamplerFactory.CreateMultiRangeUniformSampler(intervals, seed); + } + else{ + Debug.LogWarning("EnvironmentParametersChannel received an unknown data type."); + } + m_Parameters[key] = sampler; + } else { Debug.LogWarning("EnvironmentParametersChannel received an unknown data type."); @@ -63,9 +114,9 @@ protected override void OnMessageReceived(IncomingMessage msg) /// public float GetWithDefault(string key, float defaultValue) { - float valueOut; + Func valueOut; bool hasKey = m_Parameters.TryGetValue(key, out valueOut); - return hasKey ? valueOut : defaultValue; + return hasKey ? valueOut.Invoke() : defaultValue; } /// diff --git a/com.unity.ml-agents/Tests/Editor/SamplerTests.cs b/com.unity.ml-agents/Tests/Editor/SamplerTests.cs new file mode 100644 index 0000000000..14307e6733 --- /dev/null +++ b/com.unity.ml-agents/Tests/Editor/SamplerTests.cs @@ -0,0 +1,109 @@ +using System; +using NUnit.Framework; +using System.IO; +using System.Collections.Generic; +using UnityEngine; +using Unity.MLAgents.SideChannels; + +namespace Unity.MLAgents.Tests +{ + public class SamplerTests + { + const int k_Seed = 1337; + const double k_Epsilon = 0.0001; + EnvironmentParametersChannel m_Channel; + + public SamplerTests() + { + m_Channel = SideChannelsManager.GetSideChannel(); + // if running test on its own + if (m_Channel == null) + { + m_Channel = new EnvironmentParametersChannel(); + SideChannelsManager.RegisterSideChannel(m_Channel); + } + } + [Test] + public void UniformSamplerTest() + { + float min_value = 1.0f; + float max_value = 2.0f; + string parameter = "parameter1"; + using (var outgoingMsg = new OutgoingMessage()) + { + outgoingMsg.WriteString(parameter); + // 1 indicates this meessage is a Sampler + outgoingMsg.WriteInt32(1); + outgoingMsg.WriteInt32(k_Seed); + outgoingMsg.WriteInt32((int)SamplerType.Uniform); + outgoingMsg.WriteFloat32(min_value); + outgoingMsg.WriteFloat32(max_value); + byte[] message = GetByteMessage(m_Channel, outgoingMsg); + SideChannelsManager.ProcessSideChannelData(message); + } + Assert.AreEqual(1.208888f, m_Channel.GetWithDefault(parameter, 1.0f), k_Epsilon); + Assert.AreEqual(1.118017f, m_Channel.GetWithDefault(parameter, 1.0f), k_Epsilon); + } + + [Test] + public void GaussianSamplerTest() + { + float mean = 3.0f; + float stddev = 0.2f; + string parameter = "parameter2"; + using (var outgoingMsg = new OutgoingMessage()) + { + outgoingMsg.WriteString(parameter); + // 1 indicates this meessage is a Sampler + outgoingMsg.WriteInt32(1); + outgoingMsg.WriteInt32(k_Seed); + outgoingMsg.WriteInt32((int)SamplerType.Gaussian); + outgoingMsg.WriteFloat32(mean); + outgoingMsg.WriteFloat32(stddev); + byte[] message = GetByteMessage(m_Channel, outgoingMsg); + SideChannelsManager.ProcessSideChannelData(message); + } + Assert.AreEqual(2.936162f, m_Channel.GetWithDefault(parameter, 1.0f), k_Epsilon); + Assert.AreEqual(2.951348f, m_Channel.GetWithDefault(parameter, 1.0f), k_Epsilon); + } + + [Test] + public void MultiRangeUniformSamplerTest() + { + float[] intervals = new float[4]; + intervals[0] = 1.2f; + intervals[1] = 2f; + intervals[2] = 3.2f; + intervals[3] = 4.1f; + string parameter = "parameter3"; + using (var outgoingMsg = new OutgoingMessage()) + { + outgoingMsg.WriteString(parameter); + // 1 indicates this meessage is a Sampler + outgoingMsg.WriteInt32(1); + outgoingMsg.WriteInt32(k_Seed); + outgoingMsg.WriteInt32((int)SamplerType.MultiRangeUniform); + outgoingMsg.WriteFloatList(intervals); + byte[] message = GetByteMessage(m_Channel, outgoingMsg); + SideChannelsManager.ProcessSideChannelData(message); + } + Assert.AreEqual(3.387999f, m_Channel.GetWithDefault(parameter, 1.0f), k_Epsilon); + Assert.AreEqual(1.294413f, m_Channel.GetWithDefault(parameter, 1.0f), k_Epsilon); + } + + internal static byte[] GetByteMessage(SideChannel sideChannel, OutgoingMessage msg) + { + byte[] message = msg.ToByteArray(); + using (var memStream = new MemoryStream()) + { + using (var binaryWriter = new BinaryWriter(memStream)) + { + binaryWriter.Write(sideChannel.ChannelId.ToByteArray()); + binaryWriter.Write(message.Length); + binaryWriter.Write(message); + } + return memStream.ToArray(); + } + } + } +} diff --git a/com.unity.ml-agents/Tests/Editor/SamplerTests.cs.meta b/com.unity.ml-agents/Tests/Editor/SamplerTests.cs.meta new file mode 100644 index 0000000000..ef0d54e72a --- /dev/null +++ b/com.unity.ml-agents/Tests/Editor/SamplerTests.cs.meta @@ -0,0 +1,11 @@ +fileFormatVersion: 2 +guid: 7e6609c51018d4132beda8ddedd46d91 +MonoImporter: + externalObjects: {} + serializedVersion: 2 + defaultReferences: [] + executionOrder: 0 + icon: {instanceID: 0} + userData: + assetBundleName: + assetBundleVariant: diff --git a/config/ppo/3DBall_randomize.yaml b/config/ppo/3DBall_randomize.yaml index b3c6c13f21..2f3608b880 100644 --- a/config/ppo/3DBall_randomize.yaml +++ b/config/ppo/3DBall_randomize.yaml @@ -26,16 +26,13 @@ behaviors: threaded: true parameter_randomization: - resampling-interval: 5000 mass: - sampler-type: uniform - min_value: 0.5 - max_value: 10 - gravity: - sampler-type: uniform - min_value: 7 - max_value: 12 + sampler_type: uniform + sampler_parameters: + min_value: 0.5 + max_value: 10 scale: - sampler-type: uniform - min_value: 0.75 - max_value: 3 + sampler_type: uniform + sampler_parameters: + min_value: 0.75 + max_value: 3 diff --git a/docs/Training-ML-Agents.md b/docs/Training-ML-Agents.md index 2f93938466..b61fc3b24b 100644 --- a/docs/Training-ML-Agents.md +++ b/docs/Training-ML-Agents.md @@ -435,97 +435,57 @@ behaviors: # < Same as above> parameter_randomization: - resampling-interval: 5000 mass: - sampler-type: "uniform" - min_value: 0.5 - max_value: 10 + sampler_type: uniform + sampler_parameters: + min_value: 0.5 + max_value: 10 - gravity: - sampler-type: "multirange_uniform" - intervals: [[7, 10], [15, 20]] + length: + sampler_type: multirangeuniform + sampler_parameters: + intervals: [[7, 10], [15, 20]] scale: - sampler-type: "uniform" - min_value: 0.75 - max_value: 3 + sampler_type: gaussian + sampler_parameters: + mean: 2 + st_dev: .3 ``` -Note that `mass`, `gravity` and `scale` are the names of the environment -parameters that will be sampled. If a parameter specified in the file doesn't -exist in the environment, then this parameter will be ignored. +Note that `mass`, `length` and `scale` are the names of the environment +parameters that will be sampled. These are used as keys by the `EnvironmentParameter` +class to sample new parameters via the function `GetWithDefault`. | **Setting** | **Description** | | :--------------------------- | :-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -| `resampling-interval` | Number of steps for the agent to train under a particular environment configuration before resetting the environment with a new sample of `Environment Parameters`. | -| `sampler-type` | Type of sampler use for this `Environment Parameter`. This is a string that should exist in the `Sampler Factory` (explained below). | -| `sampler-type-sub-arguments` | Specify the sub-arguments depending on the `sampler-type`. In the example above, this would correspond to the `intervals` under the `sampler-type` `multirange_uniform` for the `Environment Parameter` called `gravity`. The key name should match the name of the corresponding argument in the sampler definition (explained) below) | +| `sampler_type` | A string identifier for the type of sampler to use for this `Environment Parameter`. | +| `sampler_parameters` | The parameters for a given `sampler_type`. Samplers of different types can have different `sampler_parameters` | -#### Included Sampler Types +#### Supported Sampler Types -Below is a list of included `sampler-type` as part of the toolkit. +Below is a list of the `sampler_type` values supported by the toolkit. - `uniform` - Uniform sampler - - Uniformly samples a single float value between defined endpoints. The - sub-arguments for this sampler to specify the interval endpoints are as - below. The sampling is done in the range of [`min_value`, `max_value`). - - **sub-arguments** - `min_value`, `max_value` + - Uniformly samples a single float value from a range with a given minimum + and maximum value (inclusive). + - **parameters** - `min_value`, `max_value` - `gaussian` - Gaussian sampler - - Samples a single float value from the distribution characterized by the mean - and standard deviation. The sub-arguments to specify the Gaussian - distribution to use are as below. - - **sub-arguments** - `mean`, `st_dev` + - Samples a single float value from a normal distribution with a given mean + and standard deviation. + - **parameters** - `mean`, `st_dev` - `multirange_uniform` - Multirange uniform sampler - - Uniformly samples a single float value between the specified intervals. - Samples by first performing a weight pick of an interval from the list of - intervals (weighted based on interval width) and samples uniformly from the - selected interval (half-closed interval, same as the uniform sampler). This - sampler can take an arbitrary number of intervals in a list in the following - format: [[`interval_1_min`, `interval_1_max`], [`interval_2_min`, + - First, samples an interval from a set of intervals in proportion to relative + length of the intervals. Then, uniformly samples a single float value from the + sampled interval (inclusive). This sampler can take an arbitrary number of + intervals in a list in the following format: + [[`interval_1_min`, `interval_1_max`], [`interval_2_min`, `interval_2_max`], ...] - - **sub-arguments** - `intervals` + - **parameters** - `intervals` The implementation of the samplers can be found in the -[sampler_class.py file](../ml-agents/mlagents/trainers/sampler_class.py). - -#### Defining a New Sampler Type - -If you want to define your own sampler type, you must first inherit the -_Sampler_ base class (included in the `sampler_class` file) and preserve the -interface. Once the class for the required method is specified, it must be -registered in the Sampler Factory. - -This can be done by subscribing to the _register_sampler_ method of the -`SamplerFactory`. The command is as follows: - -`SamplerFactory.register_sampler(*custom_sampler_string_key*, *custom_sampler_object*)` - -Once the Sampler Factory reflects the new register, the new sampler type can be -used for sample any `Environment Parameter`. For example, lets say a new sampler -type was implemented as below and we register the `CustomSampler` class with the -string `custom-sampler` in the Sampler Factory. - -```python -class CustomSampler(Sampler): - - def __init__(self, argA, argB, argC): - self.possible_vals = [argA, argB, argC] - - def sample_all(self): - return np.random.choice(self.possible_vals) -``` - -Now we need to specify the new sampler type in the sampler YAML file. For -example, we use this new sampler type for the `Environment Parameter` _mass_. - -```yaml -mass: - sampler-type: "custom-sampler" - argB: 1 - argA: 2 - argC: 3 -``` +[Samplers.cs file](../com.unity.ml-agents/Runtime/Sampler.cs). #### Training with Environment Parameter Randomization diff --git a/ml-agents-envs/mlagents_envs/side_channel/environment_parameters_channel.py b/ml-agents-envs/mlagents_envs/side_channel/environment_parameters_channel.py index 958364b675..2d379cbb3f 100644 --- a/ml-agents-envs/mlagents_envs/side_channel/environment_parameters_channel.py +++ b/ml-agents-envs/mlagents_envs/side_channel/environment_parameters_channel.py @@ -2,6 +2,7 @@ from mlagents_envs.exception import UnityCommunicationException import uuid from enum import IntEnum +from typing import List, Tuple class EnvironmentParametersChannel(SideChannel): @@ -13,6 +14,12 @@ class EnvironmentParametersChannel(SideChannel): class EnvironmentDataTypes(IntEnum): FLOAT = 0 + SAMPLER = 1 + + class SamplerTypes(IntEnum): + UNIFORM = 0 + GAUSSIAN = 1 + MULTIRANGEUNIFORM = 2 def __init__(self) -> None: channel_id = uuid.UUID(("534c891e-810f-11ea-a9d0-822485860400")) @@ -35,3 +42,59 @@ def set_float_parameter(self, key: str, value: float) -> None: msg.write_int32(self.EnvironmentDataTypes.FLOAT) msg.write_float32(value) super().queue_message_to_send(msg) + + def set_uniform_sampler_parameters( + self, key: str, min_value: float, max_value: float, seed: int + ) -> None: + """ + Sets a uniform environment parameter sampler. + :param key: The string identifier of the parameter. + :param min_value: The minimum of the sampling distribution. + :param max_value: The maximum of the sampling distribution. + :param seed: The random seed to initialize the sampler. + """ + msg = OutgoingMessage() + msg.write_string(key) + msg.write_int32(self.EnvironmentDataTypes.SAMPLER) + msg.write_int32(seed) + msg.write_int32(self.SamplerTypes.UNIFORM) + msg.write_float32(min_value) + msg.write_float32(max_value) + super().queue_message_to_send(msg) + + def set_gaussian_sampler_parameters( + self, key: str, mean: float, st_dev: float, seed: int + ) -> None: + """ + Sets a gaussian environment parameter sampler. + :param key: The string identifier of the parameter. + :param mean: The mean of the sampling distribution. + :param st_dev: The standard deviation of the sampling distribution. + :param seed: The random seed to initialize the sampler. + """ + msg = OutgoingMessage() + msg.write_string(key) + msg.write_int32(self.EnvironmentDataTypes.SAMPLER) + msg.write_int32(seed) + msg.write_int32(self.SamplerTypes.GAUSSIAN) + msg.write_float32(mean) + msg.write_float32(st_dev) + super().queue_message_to_send(msg) + + def set_multirangeuniform_sampler_parameters( + self, key: str, intervals: List[Tuple[float, float]], seed: int + ) -> None: + """ + Sets a multirangeuniform environment parameter sampler. + :param key: The string identifier of the parameter. + :param intervals: The lists of min and max that define each uniform distribution. + :param seed: The random seed to initialize the sampler. + """ + msg = OutgoingMessage() + msg.write_string(key) + msg.write_int32(self.EnvironmentDataTypes.SAMPLER) + msg.write_int32(seed) + msg.write_int32(self.SamplerTypes.MULTIRANGEUNIFORM) + flattened_intervals = [value for interval in intervals for value in interval] + msg.write_float32_list(flattened_intervals) + super().queue_message_to_send(msg) diff --git a/ml-agents/mlagents/trainers/learn.py b/ml-agents/mlagents/trainers/learn.py index 33c2c72b33..ead4b92cd8 100644 --- a/ml-agents/mlagents/trainers/learn.py +++ b/ml-agents/mlagents/trainers/learn.py @@ -22,8 +22,6 @@ ) from mlagents.trainers.cli_utils import parser from mlagents_envs.environment import UnityEnvironment -from mlagents.trainers.sampler_class import SamplerManager -from mlagents.trainers.exception import SamplerException from mlagents.trainers.settings import RunOptions from mlagents.trainers.training_status import GlobalTrainingStatus from mlagents_envs.base_env import BaseEnv @@ -133,9 +131,7 @@ def run_training(run_seed: int, options: RunOptions) -> None: maybe_meta_curriculum = try_create_meta_curriculum( options.curriculum, env_manager, restore=checkpoint_settings.resume ) - sampler_manager, resampling_interval = create_sampler_manager( - options.parameter_randomization, run_seed - ) + maybe_add_samplers(options.parameter_randomization, env_manager, run_seed) trainer_factory = TrainerFactory( options.behaviors, write_path, @@ -154,8 +150,6 @@ def run_training(run_seed: int, options: RunOptions) -> None: maybe_meta_curriculum, not checkpoint_settings.inference, run_seed, - sampler_manager, - resampling_interval, ) # Begin training @@ -197,26 +191,21 @@ def write_timing_tree(output_dir: str) -> None: ) -def create_sampler_manager(sampler_config, run_seed=None): - resample_interval = None +def maybe_add_samplers( + sampler_config: Optional[Dict], env: SubprocessEnvManager, run_seed: int +) -> None: + """ + Adds samplers to env if sampler config provided and sets seed if not configured. + :param sampler_config: validated dict of sampler configs. None if not included. + :param env: env manager to pass samplers via reset + :param run_seed: Random seed used for training. + """ if sampler_config is not None: - if "resampling-interval" in sampler_config: - # Filter arguments that do not exist in the environment - resample_interval = sampler_config.pop("resampling-interval") - if (resample_interval <= 0) or (not isinstance(resample_interval, int)): - raise SamplerException( - "Specified resampling-interval is not valid. Please provide" - " a positive integer value for resampling-interval" - ) - - else: - raise SamplerException( - "Resampling interval was not specified in the sampler file." - " Please specify it with the 'resampling-interval' key in the sampler config file." - ) - - sampler_manager = SamplerManager(sampler_config, run_seed) - return sampler_manager, resample_interval + # If the seed is not specified in yaml, this will grab the run seed + for offset, v in enumerate(sampler_config.values()): + if v.seed == -1: + v.seed = run_seed + offset + env.reset(config=sampler_config) def try_create_meta_curriculum( diff --git a/ml-agents/mlagents/trainers/sampler_class.py b/ml-agents/mlagents/trainers/sampler_class.py deleted file mode 100644 index f1a7c20327..0000000000 --- a/ml-agents/mlagents/trainers/sampler_class.py +++ /dev/null @@ -1,193 +0,0 @@ -import numpy as np -from typing import Union, Optional, Type, List, Dict, Any -from abc import ABC, abstractmethod - -from mlagents.trainers.exception import SamplerException - - -class Sampler(ABC): - @abstractmethod - def sample_parameter(self) -> float: - pass - - -class UniformSampler(Sampler): - """ - Uniformly draws a single sample in the range [min_value, max_value). - """ - - def __init__( - self, - min_value: Union[int, float], - max_value: Union[int, float], - seed: Optional[int] = None, - ): - """ - :param min_value: minimum value of the range to be sampled uniformly from - :param max_value: maximum value of the range to be sampled uniformly from - :param seed: Random seed used for making draws from the uniform sampler - """ - self.min_value = min_value - self.max_value = max_value - # Draw from random state to allow for consistent reset parameter draw for a seed - self.random_state = np.random.RandomState(seed) - - def sample_parameter(self) -> float: - """ - Draws and returns a sample from the specified interval - """ - return self.random_state.uniform(self.min_value, self.max_value) - - -class MultiRangeUniformSampler(Sampler): - """ - Draws a single sample uniformly from the intervals provided. The sampler - first picks an interval based on a weighted selection, with the weights - assigned to an interval based on its range. After picking the range, - it proceeds to pick a value uniformly in that range. - """ - - def __init__( - self, intervals: List[List[Union[int, float]]], seed: Optional[int] = None - ): - """ - :param intervals: List of intervals to draw uniform samples from - :param seed: Random seed used for making uniform draws from the specified intervals - """ - self.intervals = intervals - # Measure the length of the intervals - interval_lengths = [abs(x[1] - x[0]) for x in self.intervals] - cum_interval_length = sum(interval_lengths) - # Assign weights to an interval proportionate to the interval size - self.interval_weights = [x / cum_interval_length for x in interval_lengths] - # Draw from random state to allow for consistent reset parameter draw for a seed - self.random_state = np.random.RandomState(seed) - - def sample_parameter(self) -> float: - """ - Selects an interval to pick and then draws a uniform sample from the picked interval - """ - cur_min, cur_max = self.intervals[ - self.random_state.choice(len(self.intervals), p=self.interval_weights) - ] - return self.random_state.uniform(cur_min, cur_max) - - -class GaussianSampler(Sampler): - """ - Draw a single sample value from a normal (gaussian) distribution. - This sampler is characterized by the mean and the standard deviation. - """ - - def __init__( - self, - mean: Union[float, int], - st_dev: Union[float, int], - seed: Optional[int] = None, - ): - """ - :param mean: Specifies the mean of the gaussian distribution to draw from - :param st_dev: Specifies the standard devation of the gaussian distribution to draw from - :param seed: Random seed used for making gaussian draws from the sample - """ - self.mean = mean - self.st_dev = st_dev - # Draw from random state to allow for consistent reset parameter draw for a seed - self.random_state = np.random.RandomState(seed) - - def sample_parameter(self) -> float: - """ - Returns a draw from the specified Gaussian distribution - """ - return self.random_state.normal(self.mean, self.st_dev) - - -class SamplerFactory: - """ - Maintain a directory of all samplers available. - Add new samplers using the register_sampler method. - """ - - NAME_TO_CLASS = { - "uniform": UniformSampler, - "gaussian": GaussianSampler, - "multirange_uniform": MultiRangeUniformSampler, - } - - @staticmethod - def register_sampler(name: str, sampler_cls: Type[Sampler]) -> None: - """ - Registers the sampe in the Sampler Factory to be used later - :param name: String name to set as key for the sampler_cls in the factory - :param sampler_cls: Sampler object to associate to the name in the factory - """ - SamplerFactory.NAME_TO_CLASS[name] = sampler_cls - - @staticmethod - def init_sampler_class( - name: str, params: Dict[str, Any], seed: Optional[int] = None - ) -> Sampler: - """ - Initializes the sampler class associated with the name with the params - :param name: Name of the sampler in the factory to initialize - :param params: Parameters associated to the sampler attached to the name - :param seed: Random seed to be used to set deterministic random draws for the sampler - """ - if name not in SamplerFactory.NAME_TO_CLASS: - raise SamplerException( - name + " sampler is not registered in the SamplerFactory." - " Use the register_sample method to register the string" - " associated to your sampler in the SamplerFactory." - ) - sampler_cls = SamplerFactory.NAME_TO_CLASS[name] - params["seed"] = seed - try: - return sampler_cls(**params) - except TypeError: - raise SamplerException( - "The sampler class associated to the " + name + " key in the factory " - "was not provided the required arguments. Please ensure that the sampler " - "config file consists of the appropriate keys for this sampler class." - ) - - -class SamplerManager: - def __init__( - self, reset_param_dict: Dict[str, Any], seed: Optional[int] = None - ) -> None: - """ - :param reset_param_dict: Arguments needed for initializing the samplers - :param seed: Random seed to be used for drawing samples from the samplers - """ - self.reset_param_dict = reset_param_dict if reset_param_dict else {} - assert isinstance(self.reset_param_dict, dict) - self.samplers: Dict[str, Sampler] = {} - for param_name, cur_param_dict in self.reset_param_dict.items(): - if "sampler-type" not in cur_param_dict: - raise SamplerException( - "'sampler_type' argument hasn't been supplied for the {0} parameter".format( - param_name - ) - ) - sampler_name = cur_param_dict.pop("sampler-type") - param_sampler = SamplerFactory.init_sampler_class( - sampler_name, cur_param_dict, seed - ) - - self.samplers[param_name] = param_sampler - - def is_empty(self) -> bool: - """ - Check for if sampler_manager is empty. - """ - return not bool(self.samplers) - - def sample_all(self) -> Dict[str, float]: - """ - Loop over all samplers and draw a sample from each one for generating - next set of reset parameter values. - """ - res = {} - for param_name, param_sampler in list(self.samplers.items()): - res[param_name] = param_sampler.sample_parameter() - return res diff --git a/ml-agents/mlagents/trainers/settings.py b/ml-agents/mlagents/trainers/settings.py index 5c7830e4b8..49a8e8036b 100644 --- a/ml-agents/mlagents/trainers/settings.py +++ b/ml-agents/mlagents/trainers/settings.py @@ -1,15 +1,23 @@ import attr import cattr -from typing import Dict, Optional, List, Any, DefaultDict, Mapping +from typing import Dict, Optional, List, Any, DefaultDict, Mapping, Tuple from enum import Enum import collections import argparse +import abc from mlagents.trainers.cli_utils import StoreConfigFile, DetectDefault, parser from mlagents.trainers.cli_utils import load_config from mlagents.trainers.exception import TrainerConfigError from mlagents.trainers.models import ScheduleType, EncoderType +from mlagents_envs import logging_util +from mlagents_envs.side_channel.environment_parameters_channel import ( + EnvironmentParametersChannel, +) + +logger = logging_util.get_logger(__name__) + def check_and_structure(key: str, value: Any, class_type: type) -> Any: attr_fields_dict = attr.fields_dict(class_type) @@ -151,6 +159,148 @@ class CuriositySettings(RewardSignalSettings): learning_rate: float = 3e-4 +class ParameterRandomizationType(Enum): + UNIFORM: str = "uniform" + GAUSSIAN: str = "gaussian" + MULTIRANGEUNIFORM: str = "multirangeuniform" + + def to_settings(self) -> type: + _mapping = { + ParameterRandomizationType.UNIFORM: UniformSettings, + ParameterRandomizationType.GAUSSIAN: GaussianSettings, + ParameterRandomizationType.MULTIRANGEUNIFORM: MultiRangeUniformSettings, + } + return _mapping[self] + + +@attr.s(auto_attribs=True) +class ParameterRandomizationSettings(abc.ABC): + seed: int = parser.get_default("seed") + + @staticmethod + def structure(d: Mapping, t: type) -> Any: + """ + Helper method to structure a Dict of ParameterRandomizationSettings class. Meant to be registered with + cattr.register_structure_hook() and called with cattr.structure(). This is needed to handle + the special Enum selection of ParameterRandomizationSettings classes. + """ + if not isinstance(d, Mapping): + raise TrainerConfigError( + f"Unsupported parameter randomization configuration {d}." + ) + d_final: Dict[str, List[float]] = {} + for environment_parameter, environment_parameter_config in d.items(): + if environment_parameter == "resampling-interval": + logger.warning( + "The resampling-interval is no longer necessary for parameter randomization. It is being ignored." + ) + continue + if "sampler_type" not in environment_parameter_config: + raise TrainerConfigError( + f"Sampler configuration for {environment_parameter} does not contain sampler_type." + ) + if "sampler_parameters" not in environment_parameter_config: + raise TrainerConfigError( + f"Sampler configuration for {environment_parameter} does not contain sampler_parameters." + ) + enum_key = ParameterRandomizationType( + environment_parameter_config["sampler_type"] + ) + t = enum_key.to_settings() + d_final[environment_parameter] = strict_to_cls( + environment_parameter_config["sampler_parameters"], t + ) + return d_final + + @abc.abstractmethod + def apply(self, key: str, env_channel: EnvironmentParametersChannel) -> None: + """ + Helper method to send sampler settings over EnvironmentParametersChannel + Calls the appropriate sampler type set method. + :param key: environment parameter to be sampled + :param env_channel: The EnvironmentParametersChannel to communicate sampler settings to environment + """ + pass + + +@attr.s(auto_attribs=True) +class UniformSettings(ParameterRandomizationSettings): + min_value: float = attr.ib() + max_value: float = 1.0 + + @min_value.default + def _min_value_default(self): + return 0.0 + + @min_value.validator + def _check_min_value(self, attribute, value): + if self.min_value > self.max_value: + raise TrainerConfigError( + "Minimum value is greater than maximum value in uniform sampler." + ) + + def apply(self, key: str, env_channel: EnvironmentParametersChannel) -> None: + """ + Helper method to send sampler settings over EnvironmentParametersChannel + Calls the uniform sampler type set method. + :param key: environment parameter to be sampled + :param env_channel: The EnvironmentParametersChannel to communicate sampler settings to environment + """ + env_channel.set_uniform_sampler_parameters( + key, self.min_value, self.max_value, self.seed + ) + + +@attr.s(auto_attribs=True) +class GaussianSettings(ParameterRandomizationSettings): + mean: float = 1.0 + st_dev: float = 1.0 + + def apply(self, key: str, env_channel: EnvironmentParametersChannel) -> None: + """ + Helper method to send sampler settings over EnvironmentParametersChannel + Calls the gaussian sampler type set method. + :param key: environment parameter to be sampled + :param env_channel: The EnvironmentParametersChannel to communicate sampler settings to environment + """ + env_channel.set_gaussian_sampler_parameters( + key, self.mean, self.st_dev, self.seed + ) + + +@attr.s(auto_attribs=True) +class MultiRangeUniformSettings(ParameterRandomizationSettings): + intervals: List[Tuple[float, float]] = attr.ib() + + @intervals.default + def _intervals_default(self): + return [[0.0, 1.0]] + + @intervals.validator + def _check_intervals(self, attribute, value): + for interval in self.intervals: + if len(interval) != 2: + raise TrainerConfigError( + f"The sampling interval {interval} must contain exactly two values." + ) + min_value, max_value = interval + if min_value > max_value: + raise TrainerConfigError( + f"Minimum value is greater than maximum value in interval {interval}." + ) + + def apply(self, key: str, env_channel: EnvironmentParametersChannel) -> None: + """ + Helper method to send sampler settings over EnvironmentParametersChannel + Calls the multirangeuniform sampler type set method. + :param key: environment parameter to be sampled + :param env_channel: The EnvironmentParametersChannel to communicate sampler settings to environment + """ + env_channel.set_multirangeuniform_sampler_parameters( + key, self.intervals, self.seed + ) + + @attr.s(auto_attribs=True) class SelfPlaySettings: save_steps: int = 20000 @@ -303,7 +453,7 @@ class RunOptions(ExportableSettings): ) env_settings: EnvironmentSettings = attr.ib(factory=EnvironmentSettings) engine_settings: EngineSettings = attr.ib(factory=EngineSettings) - parameter_randomization: Optional[Dict] = None + parameter_randomization: Optional[Dict[str, ParameterRandomizationSettings]] = None curriculum: Optional[Dict[str, CurriculumSettings]] = None checkpoint_settings: CheckpointSettings = attr.ib(factory=CheckpointSettings) @@ -314,6 +464,10 @@ class RunOptions(ExportableSettings): cattr.register_structure_hook(EnvironmentSettings, strict_to_cls) cattr.register_structure_hook(EngineSettings, strict_to_cls) cattr.register_structure_hook(CheckpointSettings, strict_to_cls) + cattr.register_structure_hook( + Dict[str, ParameterRandomizationSettings], + ParameterRandomizationSettings.structure, + ) cattr.register_structure_hook(CurriculumSettings, strict_to_cls) cattr.register_structure_hook(TrainerSettings, TrainerSettings.structure) cattr.register_structure_hook( diff --git a/ml-agents/mlagents/trainers/simple_env_manager.py b/ml-agents/mlagents/trainers/simple_env_manager.py index b335efd6fc..98cdfbbe99 100644 --- a/ml-agents/mlagents/trainers/simple_env_manager.py +++ b/ml-agents/mlagents/trainers/simple_env_manager.py @@ -5,6 +5,7 @@ from mlagents_envs.timers import timed from mlagents.trainers.action_info import ActionInfo from mlagents.trainers.brain import BrainParameters +from mlagents.trainers.settings import ParameterRandomizationSettings from mlagents_envs.side_channel.environment_parameters_channel import ( EnvironmentParametersChannel, ) @@ -44,7 +45,10 @@ def _reset_env( ) -> List[EnvironmentStep]: # type: ignore if config is not None: for k, v in config.items(): - self.env_params.set_float_parameter(k, v) + if isinstance(v, float): + self.env_params.set_float_parameter(k, v) + elif isinstance(v, ParameterRandomizationSettings): + v.apply(k, self.env_params) self.env.reset() all_step_result = self._generate_all_results() self.previous_step = EnvironmentStep(all_step_result, 0, {}, {}) diff --git a/ml-agents/mlagents/trainers/subprocess_env_manager.py b/ml-agents/mlagents/trainers/subprocess_env_manager.py index 0687cbdf14..8bf2e4e771 100644 --- a/ml-agents/mlagents/trainers/subprocess_env_manager.py +++ b/ml-agents/mlagents/trainers/subprocess_env_manager.py @@ -23,6 +23,7 @@ get_timer_root, ) from mlagents.trainers.brain import BrainParameters +from mlagents.trainers.settings import ParameterRandomizationSettings from mlagents.trainers.action_info import ActionInfo from mlagents_envs.side_channel.environment_parameters_channel import ( EnvironmentParametersChannel, @@ -175,7 +176,10 @@ def external_brains(): _send_response(EnvironmentCommand.EXTERNAL_BRAINS, external_brains()) elif req.cmd == EnvironmentCommand.RESET: for k, v in req.payload.items(): - env_parameters.set_float_parameter(k, v) + if isinstance(v, float): + env_parameters.set_float_parameter(k, v) + elif isinstance(v, ParameterRandomizationSettings): + v.apply(k, env_parameters) env.reset() all_step_result = _generate_all_results() _send_response(EnvironmentCommand.RESET, all_step_result) diff --git a/ml-agents/mlagents/trainers/tests/test_config_conversion.py b/ml-agents/mlagents/trainers/tests/test_config_conversion.py index 00bfc42ac0..49a1489f12 100644 --- a/ml-agents/mlagents/trainers/tests/test_config_conversion.py +++ b/ml-agents/mlagents/trainers/tests/test_config_conversion.py @@ -152,12 +152,20 @@ def test_convert_behaviors(trainer_type, use_recurrent): assert RewardSignalType.CURIOSITY in trainer_settings.reward_signals +@mock.patch("mlagents.trainers.upgrade_config.convert_samplers") @mock.patch("mlagents.trainers.upgrade_config.convert_behaviors") @mock.patch("mlagents.trainers.upgrade_config.remove_nones") @mock.patch("mlagents.trainers.upgrade_config.write_to_yaml_file") @mock.patch("mlagents.trainers.upgrade_config.parse_args") @mock.patch("mlagents.trainers.upgrade_config.load_config") -def test_main(mock_load, mock_parse, yaml_write_mock, remove_none_mock, mock_convert): +def test_main( + mock_load, + mock_parse, + yaml_write_mock, + remove_none_mock, + mock_convert_behaviors, + mock_convert_samplers, +): test_output_file = "test.yaml" mock_load.side_effect = [ yaml.safe_load(PPO_CONFIG), @@ -171,7 +179,8 @@ def test_main(mock_load, mock_parse, yaml_write_mock, remove_none_mock, mock_con sampler="test", ) mock_parse.return_value = mock_args - mock_convert.return_value = "test_converted_config" + mock_convert_behaviors.return_value = "test_converted_config" + mock_convert_samplers.return_value = "test_converted_sampler_config" dict_without_nones = mock.Mock(name="nonones") remove_none_mock.return_value = dict_without_nones @@ -181,7 +190,7 @@ def test_main(mock_load, mock_parse, yaml_write_mock, remove_none_mock, mock_con yaml_write_mock.assert_called_with(dict_without_nones, test_output_file) assert saved_dict["behaviors"] == "test_converted_config" assert saved_dict["curriculum"] == "test_curriculum_config" - assert saved_dict["parameter_randomization"] == "test_sampler_config" + assert saved_dict["parameter_randomization"] == "test_converted_sampler_config" def test_remove_nones(): diff --git a/ml-agents/mlagents/trainers/tests/test_learn.py b/ml-agents/mlagents/trainers/tests/test_learn.py index 5a67036ade..167fe157e3 100644 --- a/ml-agents/mlagents/trainers/tests/test_learn.py +++ b/ml-agents/mlagents/trainers/tests/test_learn.py @@ -7,6 +7,7 @@ from mlagents.trainers.cli_utils import DetectDefault from mlagents_envs.exception import UnityEnvironmentException from mlagents.trainers.stats import StatsReporter +from mlagents.trainers.settings import UniformSettings def basic_options(extra_args=None): @@ -45,7 +46,10 @@ def basic_options(extra_args=None): MOCK_SAMPLER_CURRICULUM_YAML = """ parameter_randomization: - sampler1: foo + sampler1: + sampler_type: uniform + sampler_parameters: + min_value: 0.2 curriculum: behavior1: @@ -61,7 +65,6 @@ def basic_options(extra_args=None): @patch("mlagents.trainers.learn.write_run_options") @patch("mlagents.trainers.learn.handle_existing_directories") @patch("mlagents.trainers.learn.TrainerFactory") -@patch("mlagents.trainers.learn.SamplerManager") @patch("mlagents.trainers.learn.SubprocessEnvManager") @patch("mlagents.trainers.learn.create_environment_factory") @patch("mlagents.trainers.settings.load_config") @@ -69,7 +72,6 @@ def test_run_training( load_config, create_environment_factory, subproc_env_mock, - sampler_manager_mock, trainer_factory_mock, handle_dir_mock, write_run_options_mock, @@ -87,14 +89,7 @@ def test_run_training( options = basic_options() learn.run_training(0, options) mock_init.assert_called_once_with( - trainer_factory_mock.return_value, - "results/ppo", - "ppo", - None, - True, - 0, - sampler_manager_mock.return_value, - None, + trainer_factory_mock.return_value, "results/ppo", "ppo", None, True, 0 ) handle_dir_mock.assert_called_once_with( "results/ppo", False, False, "results/notuselessrun" @@ -216,7 +211,7 @@ def test_yaml_args(mock_file): @patch("builtins.open", new_callable=mock_open, read_data=MOCK_SAMPLER_CURRICULUM_YAML) def test_sampler_configs(mock_file): opt = parse_command_line(["mytrainerpath"]) - assert opt.parameter_randomization == {"sampler1": "foo"} + assert isinstance(opt.parameter_randomization["sampler1"], UniformSettings) assert len(opt.curriculum.keys()) == 2 diff --git a/ml-agents/mlagents/trainers/tests/test_sampler_class.py b/ml-agents/mlagents/trainers/tests/test_sampler_class.py deleted file mode 100644 index 29954c6599..0000000000 --- a/ml-agents/mlagents/trainers/tests/test_sampler_class.py +++ /dev/null @@ -1,96 +0,0 @@ -import pytest - -from mlagents.trainers.sampler_class import SamplerManager -from mlagents.trainers.sampler_class import ( - UniformSampler, - MultiRangeUniformSampler, - GaussianSampler, -) -from mlagents.trainers.exception import TrainerError - - -def sampler_config_1(): - return { - "mass": {"sampler-type": "uniform", "min_value": 5, "max_value": 10}, - "gravity": { - "sampler-type": "multirange_uniform", - "intervals": [[8, 11], [15, 20]], - }, - } - - -def check_value_in_intervals(val, intervals): - check_in_bounds = [a <= val <= b for a, b in intervals] - return any(check_in_bounds) - - -def test_sampler_config_1(): - config = sampler_config_1() - sampler = SamplerManager(config) - - assert sampler.is_empty() is False - assert isinstance(sampler.samplers["mass"], UniformSampler) - assert isinstance(sampler.samplers["gravity"], MultiRangeUniformSampler) - - cur_sample = sampler.sample_all() - - # Check uniform sampler for mass - assert sampler.samplers["mass"].min_value == config["mass"]["min_value"] - assert sampler.samplers["mass"].max_value == config["mass"]["max_value"] - assert config["mass"]["min_value"] <= cur_sample["mass"] - assert config["mass"]["max_value"] >= cur_sample["mass"] - - # Check multirange_uniform sampler for gravity - assert sampler.samplers["gravity"].intervals == config["gravity"]["intervals"] - assert check_value_in_intervals( - cur_sample["gravity"], sampler.samplers["gravity"].intervals - ) - - -def sampler_config_2(): - return {"angle": {"sampler-type": "gaussian", "mean": 0, "st_dev": 1}} - - -def test_sampler_config_2(): - config = sampler_config_2() - sampler = SamplerManager(config) - assert sampler.is_empty() is False - assert isinstance(sampler.samplers["angle"], GaussianSampler) - - # Check angle gaussian sampler - assert sampler.samplers["angle"].mean == config["angle"]["mean"] - assert sampler.samplers["angle"].st_dev == config["angle"]["st_dev"] - - -def test_empty_samplers(): - empty_sampler = SamplerManager({}) - assert empty_sampler.is_empty() - empty_cur_sample = empty_sampler.sample_all() - assert empty_cur_sample == {} - - none_sampler = SamplerManager(None) - assert none_sampler.is_empty() - none_cur_sample = none_sampler.sample_all() - assert none_cur_sample == {} - - -def incorrect_uniform_sampler(): - # Do not specify required arguments to uniform sampler - return {"mass": {"sampler-type": "uniform", "min-value": 10}} - - -def incorrect_sampler_config(): - # Do not specify 'sampler-type' key - return {"mass": {"min-value": 2, "max-value": 30}} - - -def test_incorrect_uniform_sampler(): - config = incorrect_uniform_sampler() - with pytest.raises(TrainerError): - SamplerManager(config) - - -def test_incorrect_sampler(): - config = incorrect_sampler_config() - with pytest.raises(TrainerError): - SamplerManager(config) diff --git a/ml-agents/mlagents/trainers/tests/test_settings.py b/ml-agents/mlagents/trainers/tests/test_settings.py index 6a8b2b9355..14928599fa 100644 --- a/ml-agents/mlagents/trainers/tests/test_settings.py +++ b/ml-agents/mlagents/trainers/tests/test_settings.py @@ -11,6 +11,10 @@ RewardSignalType, RewardSignalSettings, CuriositySettings, + ParameterRandomizationSettings, + UniformSettings, + GaussianSettings, + MultiRangeUniformSettings, TrainerType, strict_to_cls, ) @@ -149,3 +153,85 @@ def test_reward_signal_structure(): RewardSignalSettings.structure( "notadict", Dict[RewardSignalType, RewardSignalSettings] ) + + +def test_parameter_randomization_structure(): + """ + Tests the ParameterRandomizationSettings structure method and all validators. + """ + parameter_randomization_dict = { + "mass": { + "sampler_type": "uniform", + "sampler_parameters": {"min_value": 1.0, "max_value": 2.0}, + }, + "scale": { + "sampler_type": "gaussian", + "sampler_parameters": {"mean": 1.0, "st_dev": 2.0}, + }, + "length": { + "sampler_type": "multirangeuniform", + "sampler_parameters": {"intervals": [[1.0, 2.0], [3.0, 4.0]]}, + }, + } + parameter_randomization_distributions = ParameterRandomizationSettings.structure( + parameter_randomization_dict, Dict[str, ParameterRandomizationSettings] + ) + assert isinstance(parameter_randomization_distributions["mass"], UniformSettings) + assert isinstance(parameter_randomization_distributions["scale"], GaussianSettings) + assert isinstance( + parameter_randomization_distributions["length"], MultiRangeUniformSettings + ) + + # Check invalid distribution type + invalid_distribution_dict = { + "mass": { + "sampler_type": "beta", + "sampler_parameters": {"alpha": 1.0, "beta": 2.0}, + } + } + with pytest.raises(ValueError): + ParameterRandomizationSettings.structure( + invalid_distribution_dict, Dict[str, ParameterRandomizationSettings] + ) + + # Check min less than max in uniform + invalid_distribution_dict = { + "mass": { + "sampler_type": "uniform", + "sampler_parameters": {"min_value": 2.0, "max_value": 1.0}, + } + } + with pytest.raises(TrainerConfigError): + ParameterRandomizationSettings.structure( + invalid_distribution_dict, Dict[str, ParameterRandomizationSettings] + ) + + # Check min less than max in multirange + invalid_distribution_dict = { + "mass": { + "sampler_type": "multirangeuniform", + "sampler_parameters": {"intervals": [[2.0, 1.0]]}, + } + } + with pytest.raises(TrainerConfigError): + ParameterRandomizationSettings.structure( + invalid_distribution_dict, Dict[str, ParameterRandomizationSettings] + ) + + # Check multirange has valid intervals + invalid_distribution_dict = { + "mass": { + "sampler_type": "multirangeuniform", + "sampler_parameters": {"intervals": [[1.0, 2.0], [3.0]]}, + } + } + with pytest.raises(TrainerConfigError): + ParameterRandomizationSettings.structure( + invalid_distribution_dict, Dict[str, ParameterRandomizationSettings] + ) + + # Check non-Dict input + with pytest.raises(TrainerConfigError): + ParameterRandomizationSettings.structure( + "notadict", Dict[str, ParameterRandomizationSettings] + ) diff --git a/ml-agents/mlagents/trainers/tests/test_simple_rl.py b/ml-agents/mlagents/trainers/tests/test_simple_rl.py index 0407601957..67d3c66617 100644 --- a/ml-agents/mlagents/trainers/tests/test_simple_rl.py +++ b/ml-agents/mlagents/trainers/tests/test_simple_rl.py @@ -13,7 +13,6 @@ from mlagents.trainers.trainer_controller import TrainerController from mlagents.trainers.trainer_util import TrainerFactory from mlagents.trainers.simple_env_manager import SimpleEnvManager -from mlagents.trainers.sampler_class import SamplerManager from mlagents.trainers.demo_loader import write_demo from mlagents.trainers.stats import StatsReporter, StatsWriter, StatsSummary from mlagents.trainers.settings import ( @@ -138,8 +137,6 @@ def _check_environment_trains( meta_curriculum=meta_curriculum, train=True, training_seed=seed, - sampler_manager=SamplerManager(None), - resampling_interval=None, ) # Begin training diff --git a/ml-agents/mlagents/trainers/tests/test_trainer_controller.py b/ml-agents/mlagents/trainers/tests/test_trainer_controller.py index cf2f872531..8a0280dc4b 100644 --- a/ml-agents/mlagents/trainers/tests/test_trainer_controller.py +++ b/ml-agents/mlagents/trainers/tests/test_trainer_controller.py @@ -4,7 +4,6 @@ from mlagents.tf_utils import tf from mlagents.trainers.trainer_controller import TrainerController from mlagents.trainers.ghost.controller import GhostController -from mlagents.trainers.sampler_class import SamplerManager @pytest.fixture @@ -18,8 +17,6 @@ def basic_trainer_controller(): meta_curriculum=None, train=True, training_seed=99, - sampler_manager=SamplerManager({}), - resampling_interval=None, ) @@ -36,8 +33,6 @@ def test_initialization_seed(numpy_random_seed, tensorflow_set_seed): meta_curriculum=None, train=True, training_seed=seed, - sampler_manager=SamplerManager({}), - resampling_interval=None, ) numpy_random_seed.assert_called_with(seed) tensorflow_set_seed.assert_called_with(seed) diff --git a/ml-agents/mlagents/trainers/trainer_controller.py b/ml-agents/mlagents/trainers/trainer_controller.py index 3a8a74d15f..fdb73bca03 100644 --- a/ml-agents/mlagents/trainers/trainer_controller.py +++ b/ml-agents/mlagents/trainers/trainer_controller.py @@ -17,7 +17,6 @@ UnityCommunicationException, UnityCommunicatorStoppedException, ) -from mlagents.trainers.sampler_class import SamplerManager from mlagents_envs.timers import ( hierarchical_timer, timed, @@ -42,8 +41,6 @@ def __init__( meta_curriculum: Optional[MetaCurriculum], train: bool, training_seed: int, - sampler_manager: SamplerManager, - resampling_interval: Optional[int], ): """ :param output_path: Path to save the model. @@ -52,8 +49,6 @@ def __init__( :param meta_curriculum: MetaCurriculum object which stores information about all curricula. :param train: Whether to train model, or only run inference. :param training_seed: Seed to use for Numpy and Tensorflow random number generation. - :param sampler_manager: SamplerManager object handles samplers for resampling the reset parameters. - :param resampling_interval: Specifies number of simulation steps after which reset parameters are resampled. :param threaded: Whether or not to run trainers in a separate thread. Disable for testing/debugging. """ self.trainers: Dict[str, Trainer] = {} @@ -64,8 +59,6 @@ def __init__( self.run_id = run_id self.train_model = train self.meta_curriculum = meta_curriculum - self.sampler_manager = sampler_manager - self.resampling_interval = resampling_interval self.ghost_controller = self.trainer_factory.ghost_controller self.trainer_threads: List[threading.Thread] = [] @@ -142,12 +135,10 @@ def _reset_env(self, env: EnvManager) -> None: A Data structure corresponding to the initial reset state of the environment. """ - sampled_reset_param = self.sampler_manager.sample_all() new_meta_curriculum_config = ( self.meta_curriculum.get_config() if self.meta_curriculum else {} ) - sampled_reset_param.update(new_meta_curriculum_config) - env.reset(config=sampled_reset_param) + env.reset(config=new_meta_curriculum_config) def _not_done_training(self) -> bool: return ( @@ -207,7 +198,6 @@ def _create_trainers_and_managers( def start_learning(self, env_manager: EnvManager) -> None: self._create_output_path(self.output_path) tf.reset_default_graph() - global_step = 0 last_brain_behavior_ids: Set[str] = set() try: # Initial reset @@ -219,8 +209,7 @@ def start_learning(self, env_manager: EnvManager) -> None: last_brain_behavior_ids = external_brain_behavior_ids n_steps = self.advance(env_manager) for _ in range(n_steps): - global_step += 1 - self.reset_env_if_ready(env_manager, global_step) + self.reset_env_if_ready(env_manager) # Stop advancing trainers self.join_threads() except ( @@ -258,7 +247,7 @@ def end_trainer_episodes( if changed: self.trainers[brain_name].reward_buffer.clear() - def reset_env_if_ready(self, env: EnvManager, steps: int) -> None: + def reset_env_if_ready(self, env: EnvManager) -> None: if self.meta_curriculum: # Get the sizes of the reward buffers. reward_buff_sizes = { @@ -274,16 +263,9 @@ def reset_env_if_ready(self, env: EnvManager, steps: int) -> None: # If any lessons were incremented or the environment is # ready to be reset meta_curriculum_reset = any(lessons_incremented.values()) - # Check if we are performing generalization training and we have finished the - # specified number of steps for the lesson - generalization_reset = ( - not self.sampler_manager.is_empty() - and (steps != 0) - and (self.resampling_interval) - and (steps % self.resampling_interval == 0) - ) + # If ghost trainer swapped teams ghost_controller_reset = self.ghost_controller.should_reset() - if meta_curriculum_reset or generalization_reset or ghost_controller_reset: + if meta_curriculum_reset or ghost_controller_reset: self.end_trainer_episodes(env, lessons_incremented) @timed diff --git a/ml-agents/mlagents/trainers/upgrade_config.py b/ml-agents/mlagents/trainers/upgrade_config.py index 7425708ddb..4263e6cf1f 100644 --- a/ml-agents/mlagents/trainers/upgrade_config.py +++ b/ml-agents/mlagents/trainers/upgrade_config.py @@ -82,6 +82,23 @@ def remove_nones(config: Dict[Any, Any]) -> Dict[str, Any]: return new_config +# Take a sampler from the old format and convert to new sampler structure +def convert_samplers(old_sampler_config: Dict[str, Any]) -> Dict[str, Any]: + new_sampler_config: Dict[str, Any] = {} + for parameter, parameter_config in old_sampler_config.items(): + if parameter == "resampling-interval": + print( + "resampling-interval is no longer necessary for parameter randomization and is being ignored." + ) + continue + new_sampler_config[parameter] = {} + new_sampler_config[parameter]["sampler_type"] = parameter_config["sampler-type"] + new_samp_parameters = dict(parameter_config) # Copy dict + new_samp_parameters.pop("sampler-type") + new_sampler_config[parameter]["sampler_parameters"] = new_samp_parameters + return new_sampler_config + + def parse_args(): argparser = argparse.ArgumentParser( formatter_class=argparse.ArgumentDefaultsHelpFormatter @@ -124,7 +141,8 @@ def main() -> None: full_config["curriculum"] = curriculum_config_dict if args.sampler is not None: - sampler_config_dict = load_config(args.sampler) + old_sampler_config_dict = load_config(args.sampler) + sampler_config_dict = convert_samplers(old_sampler_config_dict) full_config["parameter_randomization"] = sampler_config_dict # Convert config to dict