* reverting to older version of 3.6 python to test if the build is failing because of that
* revert 3.7 to 3.7.3 for the build
* revert python 3.8 version to 3.8.1
* do not install mujoco on 3.8 and 3.9
* enable mujoco for 3.7
* .
* .
* .
* use regex to navigate python version dependent package installation
* .
* add test showing mismatch in flattened space dtype and flattened point dtype
* fix mismatch in flattened space dtype and flattened point dtype
* fix typo
* enhance test to detect when flattened dtype is incorrect
* fix incorrect flattened dtype
* remove inaccurate comment
* change flatten to always use space.dtype
* added testing for unflattened dtypes
* fix unflatten dtypes
* swtich flatten_space to use space.dtype for hardcoded space dtypes
* fix failure in python 3.5
This env is a bit more heavyweight due to the size and complexity of the underlying KeY prover, but I hope that you too can build/test it hand have some fun there 😄
Please notify me if there is anything wrong with the installation or the env itself.
* Allow custom observation spaces in VectorEnv
* Replace np.copy by deepcopy in reset of SyncVectorEnv
* Add tests for VectorEnv with custom spaces
* Add tests for shared memory and batches of custom spaces
* Remove unused import in VectorEnv test
* Add warning note in the Space class for custom spaces
The current docstring for `reset()` seems to indicate that the
environment will be identical after separate calls to `reset()`.
However, the `reset()` function isn't supposed to reset the states of
the environment's RNGs [1]. This change clarifies the relationship
between the `reset()` function and the RNGs.
[1]: https://github.com/openai/gym/issues/250
* Update discrete.py
To make toy_text DiscreteEnv's reset() method and step(action) method return consistent type of a state.
The reset() method of DiscreteEnv returns a state of type numpy.int64, while step(action) returns a state of type int. This inconsistency could cause some problem for some agents.
In reset() method, force the returned state as type int.
* add cast to int in toy_text.discrete.step method
Co-authored-by: Peter Zhokhov <peterz@openai.com>
* include low, high, & dtype in spaces.Box.__repr__
multitask learners need a way to make a sensor for each space, and it's possible for two different Box spaces to have the same shape but different low/high/dtype, so this pull request just adds the minimum low value, maximum high value, and the dtype to Box.__repr__ so we can use str(box_space) as a key in a ModuleDict of sensors
* f-string to format
* fix typo
Co-authored-by: Peter Zhokhov <peterz@openai.com>
* Update README to reflect MuJoCo version support
an older version of mujoco-py needs to be installed with this library than what is available in master of mujoco-py repo.
* Update README to reflect MuJoCo version support
In gym.envs.classic_control.rendering, Geom has a color attribute,
which is initialized as black. This is a good default for most Geom
subclasses, such as lines and circles. However, in the case of Image
instances, the result is a completely black image - in the case of an
image without transparency, the result is a black square. This can be
easily fixed by setting the default color for Image instances as white,
which causes images to be displayed normally.
