Merge branch 'main' into coolheri

.circleci/config.yml

@@ -13,7 +13,7 @@ orbs:
 executors:
   ubuntu-python37:
     docker:
-      - image: circleci/python:3.7
+      - image: cimg/python:3.7
     working_directory: ~/repo
 
 
@@ -27,9 +27,9 @@ commands:
       - restore_cache:
           name: "[all] Restore cache"
           keys:
-          - v1-dependencies-{{ checksum "requirements/dev.txt" }}-{{ checksum "requirements/main.txt"}}-{{ checksum "requirements/bench.txt"}}
+          - v2-dependencies-{{ checksum "requirements/dev.txt" }}-{{ checksum "requirements/main.txt"}}-{{ checksum "requirements/bench.txt"}}
           # fallback to installing main requirements
-          - v1-dependencies-main-{{ checksum "requirements/main.txt"}}
+          - v2-dependencies-main-{{ checksum "requirements/main.txt"}}
 
       - run:
           name: "[all] Install dependencies"
@@ -53,7 +53,7 @@ jobs:
       - restore_cache:
           name: "[no-extra] Restore cache"
           keys:
-          - v1-dependencies-main-{{ checksum "requirements/main.txt"}}
+          - v2-dependencies-main-{{ checksum "requirements/main.txt"}}
           # fallback to using the latest cache if no exact match is found
 
       - run:
@@ -80,7 +80,7 @@ jobs:
           name: "[no-extra] Save cache"
           paths:
             - ./venv
-          key: v1-dependencies-main-{{ checksum "requirements/main.txt"}}
+          key: v2-dependencies-main-{{ checksum "requirements/main.txt"}}
 
       - run:
           name: "[no-extra] Run basic tests (checking dependencies)"
@@ -95,7 +95,7 @@ jobs:
           name: "[all] Save cache"
           paths:
             - ./venv
-          key: v1-dependencies-{{ checksum "requirements/dev.txt" }}-{{ checksum "requirements/main.txt"}}-{{ checksum "requirements/bench.txt"}}
+          key: v2-dependencies-{{ checksum "requirements/dev.txt" }}-{{ checksum "requirements/main.txt"}}-{{ checksum "requirements/bench.txt"}}
 
       - run:
           name: "[all] Print installation"
@@ -187,7 +187,7 @@ jobs:
           command: |
             . venv/bin/activate
             black --version
-            black nevergrad --check
+            black nevergrad --check --diff
 
 
   pytests:
@@ -247,7 +247,7 @@ jobs:
       - restore_cache:
           name: "[all] Restore cache"
           keys:
-          - v1-win-dependencies-{{ checksum "requirements/dev.txt" }}-{{ checksum "requirements/main.txt"}}-{{ checksum "requirements/bench.txt"}}
+          - v2-win-dependencies-{{ checksum "requirements/dev.txt" }}-{{ checksum "requirements/main.txt"}}-{{ checksum "requirements/bench.txt"}}
 
       - run:
           name: Setup nevergrad in virtualenv
@@ -262,7 +262,7 @@ jobs:
           name: "[all] Save cache"
           paths:
             - ./venv
-          key: v1-win-dependencies-{{ checksum "requirements/dev.txt" }}-{{ checksum "requirements/main.txt"}}-{{ checksum "requirements/bench.txt"}}
+          key: v2-win-dependencies-{{ checksum "requirements/dev.txt" }}-{{ checksum "requirements/main.txt"}}-{{ checksum "requirements/bench.txt"}}
 
       - run:
           name: "[all] Print installation"

CHANGELOG.md

@@ -2,6 +2,18 @@
 
 ## main
 
+### Breaking changes
+
+- Removed `descriptor` field of parameters which had been deprecated in previous versions. Use `function` field instead to specify if the function 
+  is deterministic or not [#X](https://github.com/facebookresearch/nevergrad/pull/X).
+
+### Important changes
+
+- `TransitionChoice` behavior has been changed to use bins instead of a full float representation. This may lead to slight
+  changes during optimizations. It can also be set as unordered for use with discrete 1+1 optimizers (experimental)
+
+### Other changes
+
 ## 0.5.0 (2022-03-08)
 
 ### Breaking changes

docs/machinelearning.rst

@@ -111,12 +111,12 @@ Asynchronous version with concurrent.futures
     from concurrent import futures
 
     for name in names:
-        optim = ng.optimizers.registry[name](parametrization=parametrization, budget=budget)
+        optim = ng.optimizers.registry[name](parametrization=instru, budget=budget)
 
         with futures.ThreadPoolExecutor(max_workers=optim.num_workers) as executor:  # the executor will evaluate the function in multiple threads
             recommendation = optim.minimize(train_and_return_test_error, executor=executor)
         print("* ", name, " provides a vector of parameters with test error ",
-              train_and_return_test_error(recommendation))
+              train_and_return_test_error(*recommendation.args, **recommendation.kwargs))
 
 
 Optimization of mixed (continuous and discrete) hyperparameters
@@ -184,7 +184,7 @@ Then you can run the optimization as usual. :code:`PortfolioDiscreteOnePlusOne`
     import nevergrad as ng
     budget = 1200  # How many episode we will do before concluding.
     for name in ["RandomSearch", "ScrHammersleySearch", "TwoPointsDE", "PortfolioDiscreteOnePlusOne", "CMA", "PSO"]:
-        optim = ng.optimizers.registry[name](parametrization=parametrization, budget=budget)
+        optim = ng.optimizers.registry[name](parametrization=instru, budget=budget)
         for u in range(budget // 3):
             x1 = optim.ask()
             # Ask and tell can be asynchronous.

docs/opencompetition2020.md

@@ -55,10 +55,6 @@ Make sure to select the Open Optimization Competition 2020 as track when you sub
 **Submission deadline is Sept. 30, 2020, AoE**
 All pull requests active between December 1st 2019 and Sept 30, 2020 are eligible.
 
-## Awards:
-Up to 12 000 euros of awards, to be used for traveling to PPSN or GECCO 2020 or 2021, distributed over several winners.
-In addition, a limited number of registration fee waivers are available for PPSN 2020.
-
 ## Winners:
 The winning entries of the Open Optimization Competition 2020 are
 (ordered by pull request ID):

docs/parametrization_ref.rst

@@ -9,6 +9,7 @@ The aim of parametrization is to specify what are the parameters that the optimi
 The parametrization subpackage will help you do thanks to:
 
 - the `parameter` modules (accessed by the shortcut `nevergrad.p`) providing classes that should be used to specify each parameter.
+- the `ops` module (accessed through `ng.ops`) providing experimental objects for modifying a parameter behavior (eg: casting to int, adding complex constraints).
 - the `FolderFunction` which helps transform any code into a Python function in a few lines. This can be especially helpful to optimize parameters in non-Python 3.6+ code (C++, Octave, etc...) or parameters in scripts.
 
 
@@ -39,3 +40,13 @@ Parameter API
 .. autoclass:: nevergrad.p.Parameter
     :members:
     :inherited-members:
+
+Operators
+---------
+These experimental operators are designed to be instantiated and called with a parameter as input, creating a new parameter with the required behavior.
+
+.. automodule:: nevergrad.ops
+    :members: Int
+
+
+

mypy.ini

@@ -1,9 +1,9 @@
 [mypy]
 
-[mypy-scipy.*,requests,pandas,compiler_gym,compiler_gym.*,gym,gym.*,gym_anm,matplotlib.*,pytest,cma,bayes_opt.*,torchvision.models,torch.*,mpl_toolkits.*,fcmaes.*,tqdm,pillow,PIL,PIL.Image,sklearn.*,pyomo.*,pyproj,IOHexperimenter.*,tensorflow,koncept.models,cv2,imquality,imquality.brisque,lpips,mixsimulator.*,networkx.*,cdt.*,pymoo,pymoo.*,bayes_optim.*,olympus.*,auto-sklearn.*,openml]
+[mypy-scipy.*,requests,pandas,compiler_gym,compiler_gym.*,gym_anm,,gym,gym.*,matplotlib.*,pytest,cma,bayes_opt.*,torchvision.models,torch.*,mpl_toolkits.*,fcmaes.*,tqdm,pillow,PIL,PIL.Image,sklearn.*,pyomo.*,pyproj,IOHexperimenter.*,tensorflow,koncept.models,cv2,imquality,imquality.brisque,lpips,mixsimulator.*,networkx.*,cdt.*,pymoo,pymoo.*,bayes_optim.*,olympus.*,auto-sklearn.*,openml,pymoo,pymoo.*,pybullet,pybullet_envs,pybulletgym,pyvirtualdisplay,nlopt,aquacrop.*]
 ignore_missing_imports = True
 
-[mypy-nevergrad.functions.rl.agents,torchvision,torchvision.*,nevergrad.functions.games.*,nevergrad.functions.multiobjective.pyhv,nevergrad.optimization.test_doc,,pymoo,pymoo.*,pybullet,pybullet_envs,pybulletgym,pyvirtualdisplay,nlopt,aquacrop.*]
+[mypy-nevergrad.functions.rl.*,torchvision,torchvision.*,nevergrad.functions.games.*,nevergrad.functions.multiobjective.pyhv,nevergrad.optimization.test_doc,nevergrad.functions.gym.multigym]
 ignore_missing_imports = True
 ignore_errors = True
 

nevergrad/benchmark/__main__.py

@@ -21,7 +21,7 @@ def launch(
     output: tp.Optional[tp.PathLike] = None,
 ) -> Path:
     """Launch experiment with given names and selection modulo
-    max_index can be specified to provide a limited number of settings
+    cap_index can be specified to provide a limited number of settings
     """
     # create the data
     csvpath = Path(experiment + ".csv") if output is None else Path(output)
@@ -103,7 +103,7 @@ def repeated_launch(
     repetitions: int = 1,
 ) -> None:
     """Launch experiment with given names and selection module
-    max_index can be specified to provide a limited number of settings
+    cap_index can be specified to provide a limited number of settings
     This repeats the plan several times and increments the seed.
     """
     # start by importing additional content

nevergrad/benchmark/core.py

@@ -234,7 +234,7 @@ def _submit_jobs(
     seed: int
         a seed for the experiment plan (if seedable)
     executor: Executor-like object
-        an object such as concurrent.futures.ThreadPoolExecutor for running experiments in parallel
+        an object such as concurrent.futures.ProcessPoolExecutor for running experiments in parallel
     print_function: tp.Callable
         a function to print at the end of each experiment (for custom logging)
     cap_index: int
@@ -281,7 +281,7 @@ def compute(
     seed: int
         a seed for the experiment plan (if seedable)
     executor: Executor-like object
-        an object such as concurrent.futures.ThreadPoolExecutor for running experiments in parallel
+        an object such as concurrent.futures.ProcessPoolExecutor for running experiments in parallel
     print_function: tp.Callable
         a function to print at the end of each experiment (for custom logging)
     cap_index: int