Develop generalization training

config/generalize_test.yaml

@@ -0,0 +1,16 @@
+episode-length: 5000
+
+mass:
+    sampler-type: "uniform"
+    min_value: 0.5
+    max_value: 10
+
+gravity:
+    sampler-type: "uniform"
+    min_value: 7
+    max_value: 12
+
+scale:
+    sampler-type: "uniform"
+    min_value: 0.75
+    max_value: 3

docs/Training-Generalization-Learning.md

@@ -0,0 +1,105 @@
+# Training Generalized Reinforcement Learning agents
+
+Reinforcement learning has a rather unique setup as opposed to supervised and
+unsupervised learning. Agents here are trained and tested on the same exact environment,
+which is analogous to a model being trained and tested on an identical dataset in supervised learning! This setting results in overfitting; the inability of the agent to generalize to slight tweaks or variations in the environment. This is problematic in instances when environments are randomly instantiated with varying properties. To make agents more robust, we train an agent over multiple variations of the environment. The agent is trained with the intent that it learns to maintain a minimum performance regardless of the environment variant and that it generalizes to maintain this in unseen future variants of the environment.
+
+Ball scale of 0.5          |  Ball scale of 4
+:-------------------------:|:-------------------------:
+![](images/3dball_small.png)  |  ![](images/3dball_big.png)
+
+_Variations of the 3D Ball environment._
+
+To vary the environments, we first decided what the reset parameters for the environment. These parameters are known as `Reset Parameters`, introduced in [Curriculum Learning](https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Training-Curriculum-Learning.md). In the 3D ball environment example displayed in the figure above, the reset parameters are `gravity`, `ball_mass` and `ball_scale`.
+
+## In-practice
+
+To test the effectiveness of this training procedure, we train 3 models over 50000 steps:
+    1. Model Trained on Default reset parameter values (Default model).
+    2. Model Trained on a range of reset parameter values (Random model). Reset parameter values are picked uniformly over the range and the model is trained on each configuration for 5000 steps before they are randomly sampled again. The range consists of the default values.
+    3. Model Trained on a range of reset parameters values (Extreme model). The range here are the extreme ends, which lie outside the range used to train the Random model. Reset parameter values are selected randomly and the model is trained for 5000 steps on each configuration of values before resampling the values again.
+
+Then, these 3 models are tested for their performance in all the reset parameter settings for 50000 steps each. The figure below outlines the performance of each model.
+
+![3DBall environment mean rewards](images/3DBall_generalization_rewards.png)
+
+In a relatively simple environment we are able to see differences in the performance between differently trained models, with the models exposed to various environments performing better in variations of environments than the default model trained over a single environment.
+
+## How-to
+
+For generalization training, we need to provide a way to modify the environment by supplying a set of reset parameters. This provision can be either deterministic or randomized. Each reset parameter is assigned a sampler. If a sampler isn't provided for a reset parameter, the parameter maintains the default value throughout the training, remaining unchanged. The samplers for all the reset parameters are handled by a **Sampler Manager**, which is also responsible for generating a new set of values for the reset parameters when needed. 
+
+To setup the Sampler Manager, we setup a YAML file that specifies how we wish to generate new samples. In this file, we specify the samplers and the `resampling-duration` (number of training steps after which reset parameters are resampled). Below is an example of a sampler file for the 3D ball environment.
+
+```yaml
+episode-length: 5000
+
+mass:
+    sampler-type: "uniform"
+    min_value: 0.5
+    max_value: 10
+
+gravity:
+    sampler-type: "multirange_uniform"
+    intervals: [[7, 10], [15, 20]]
+
+scale:
+    sampler-type: "uniform"
+    min_value: 0.75
+    max_value: 3
+
+```
+
+* `resampling-duration` (int) - Specifies the number of steps for agent to train under a particular environment configuration before resetting the environment with a new sample of reset parameters.
+
+* `parameter_name` - Name of the reset parameter. This should match the name specified in the academy of the intended environment for which the agent is being trained. If a parameter specified in the file doesn't exist in the environment, then this specification will be ignored.
+
+    * `sampler-type` - Specify the sampler type to use for the reset parameter. This is a string that should exist in the `Sampler Factory` (explained below).
+
+    * `sub-arguments` - Specify the characteristic parameters for the sampler. In the example sampler file above, this would correspond to the `intervals` key under the `multirange_uniform` sampler for the gravity reset parameter. The key name should match the name of the corresponding argument in the sampler definition. (Look at defining a new sampler method)
+
+The sampler manager allocates a sampler for a reset parameter by using the *Sampler Factory*, which maintains a dictionary mapping of string keys to sampler objects. The available samplers to be used for reset parameter resampling is as available in the Sampler Factory.
+
+The implementation of the samplers can be found at `ml-agents-envs/mlagents/envs/sampler_class.py`.
+
+### Defining a new sampler method
+
+Custom sampling techniques must inherit from 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 custom sampler can be used for resampling reset parameter. For demonstration, lets say our sampler 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 this sampler in the sampler file. Lets say we wish to use this sampler for the reset parameter *mass*; the sampler file would specify the same for mass as the following.
+
+```yaml
+mass:
+    sampler-type: "custom-sampler"
+    argA: 1
+    argB: 2
+    argC: 3
+```
+
+With the sampler file setup, we can proceed to train our agent as explained in the next section.
+
+### Training with Generalization Learning
+
+We first begin with setting up the sampler file. After the sampler file is defined and configured, we proceed by launching `mlagents-learn` and specify our configured sampler file with the `--sampler` flag. To demonstrate, if we wanted to train a 3D ball agent with generalization using the `generalization-test.yaml` sampling setup, we can run
+
+```sh
+mlagents-learn config/trainer_config.yaml --sampler=config/generalize_test.yaml --run-id=3D-Ball-generalization --train
+```
+
+We can observe progress and metrics via Tensorboard.

docs/Training-ML-Agents.md

@@ -101,6 +101,9 @@ environment, you can set the following command line options when invoking
 * `--curriculum=<file>` – Specify a curriculum JSON file for defining the
   lessons for curriculum training. See [Curriculum
   Training](Training-Curriculum-Learning.md) for more information.
+* `--sampler=<file>` - Specify a sampler YAML file for defining the
+  sampler for generalization training. See [Generalization
+  Training](Training-Generalization-Learning.md) for more information.
 * `--keep-checkpoints=<n>` – Specify the maximum number of model checkpoints to
   keep. Checkpoints are saved after the number of steps specified by the
   `save-freq` option. Once the maximum number of checkpoints has been reached,
@@ -197,6 +200,7 @@ are conducting, see:
 * [Training with PPO](Training-PPO.md)
 * [Using Recurrent Neural Networks](Feature-Memory.md)
 * [Training with Curriculum Learning](Training-Curriculum-Learning.md)
+* [Training with Generalization](Training-Generalization-Learning.md)
 * [Training with Imitation Learning](Training-Imitation-Learning.md)
 
 You can also compare the

ml-agents-envs/mlagents/envs/exception.py

@@ -27,6 +27,14 @@ class UnityActionException(UnityException):
     pass
 
 
+class SamplerException(UnityException):
+    """
+    Related to errors with the sampler actions.
+    """
+
+    pass
+
+
 class UnityTimeOutException(UnityException):
     """
     Related to errors with communication timeouts.

ml-agents-envs/mlagents/envs/sampler_class.py

@@ -0,0 +1,113 @@
+import numpy as np
+from typing import *
+from functools import *
+from collections import OrderedDict
+from abc import ABC, abstractmethod
+
+from .exception import SamplerException
+
+
+class Sampler(ABC):
+    @abstractmethod
+    def sample_parameter(self) -> float:
+        pass
+
+
+class UniformSampler(Sampler):
+    # kwargs acts as a sink for extra unneeded args
+    def __init__(self, min_value, max_value, **kwargs):
+        self.min_value = min_value
+        self.max_value = max_value
+
+    def sample_parameter(self) -> float:
+        return np.random.uniform(self.min_value, self.max_value)
+
+
+class MultiRangeUniformSampler(Sampler):
+    def __init__(self, intervals, **kwargs):
+        self.intervals = intervals
+        # Measure the length of the intervals
+        interval_lengths = [abs(x[1] - x[0]) for x in self.intervals]
+        # Cumulative size of the 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]
+
+    def sample_parameter(self) -> float:
+        cur_min, cur_max = self.intervals[
+            np.random.choice(len(self.intervals), p=self.interval_weights)
+        ]
+        return np.random.uniform(cur_min, cur_max)
+
+
+class GaussianSampler(Sampler):
+    def __init__(self, mean, var, **kwargs):
+        self.mean = mean
+        self.var = var
+
+    def sample_parameter(self) -> float:
+        return np.random.normal(self.mean, self.var)
+
+
+# To introduce new sampling methods, just need to 'register' them to this sampler factory
+class SamplerFactory:
+    NAME_TO_CLASS = {
+        "uniform": UniformSampler,
+        "gaussian": GaussianSampler,
+        "multirange_uniform": MultiRangeUniformSampler,
+    }
+
+    @staticmethod
+    def register_sampler(name, sampler_cls):
+        SamplerFactory.NAME_TO_CLASS[name] = sampler_cls
+
+    @staticmethod
+    def init_sampler_class(name, param_dict):
+        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]
+        try:
+            return sampler_cls(**param_dict)
+        except:
+            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):
+        self.reset_param_dict = reset_param_dict if reset_param_dict else {}
+        assert isinstance(self.reset_param_dict, dict)
+        self.samplers = OrderedDict()
+        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
+            )
+
+            self.samplers[param_name] = param_sampler
+
+    def is_empty(self) -> bool:
+        """
+        If self.samplers is empty, then bool of it returns false, indicating that the
+        sampler manager isn't managing any samplers.
+        """
+        return not bool(self.samplers)
+
+    def sample_all(self) -> Dict[str, float]:
+        res = {}
+        for param_name, param_sampler in list(self.samplers.items()):
+            res[param_name] = param_sampler.sample_parameter()
+        return res

ml-agents-envs/mlagents/envs/tests/test_sampler_class.py

@@ -0,0 +1,99 @@
+from math import isclose
+import pytest
+
+from mlagents.envs.sampler_class import SamplerManager
+from mlagents.envs.sampler_class import (
+    UniformSampler,
+    MultiRangeUniformSampler,
+    GaussianSampler,
+)
+from mlagents.envs.exception import UnityException
+
+
+def basic_3Dball_sampler():
+    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_3Dball_sampler():
+    config = basic_3Dball_sampler()
+    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 basic_tennis_sampler():
+    return {"angle": {"sampler-type": "gaussian", "mean": 0, "var": 1}}
+
+
+def test_tennis_sampler():
+    config = basic_tennis_sampler()
+    sampler = SamplerManager(config)
+    assert sampler.is_empty() is False
+    assert isinstance(sampler.samplers["angle"], GaussianSampler)
+
+    cur_sample = sampler.sample_all()
+
+    # Check angle gaussian sampler
+    assert sampler.samplers["angle"].mean == config["angle"]["mean"]
+    assert sampler.samplers["angle"].var == config["angle"]["var"]
+
+
+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(UnityException):
+        SamplerManager(config)
+
+
+def test_incorrect_sampler():
+    config = incorrect_sampler_config()
+    with pytest.raises(UnityException):
+        SamplerManager(config)

ml-agents/mlagents/trainers/exception.py

@@ -23,3 +23,5 @@ class MetaCurriculumError(TrainerError):
     """
     Any error related to the configuration of a metacurriculum.
     """
+
+    pass