merged down 2.0 to local

.github/workflows/install.yml

@@ -0,0 +1,29 @@
+name: install
+on:
+  push:
+  pull_request:
+
+jobs:
+  test:
+    name: test ${{ matrix.py }} - ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os:
+          - ubuntu-latest
+          - macos-latest
+        py:
+          - "3.11"
+          - "3.10"
+          - "3.9"
+    steps:
+      - name: Setup python for test ${{ matrix.py }}
+        uses: actions/setup-python@v4
+        with:
+          python-version: ${{ matrix.py }}
+      - uses: actions/checkout@v3
+      - name: Upgrade pip
+        run: python -m pip install -U pip
+      - name: Install pufferlib
+        run: pip3 install -e . 

.gitignore

@@ -1,3 +1,13 @@
+# Annoying temp files generated by Cython
+c_gae.c
+pufferlib/extensions.c
+pufferlib/ocean/grid/c_grid.c
+pufferlib/ocean/tactical/c_tactical.c
+pufferlib/puffernet.c
+
+# Raylib
+raylib_wasm/
+
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]

README.md

@@ -1,6 +1,8 @@
 ![figure](https://pufferai.github.io/source/resource/header.png)
 
 [![PyPI version](https://badge.fury.io/py/pufferlib.svg)](https://badge.fury.io/py/pufferlib)
+![PyPI - Python Version](https://img.shields.io/pypi/pyversions/pufferlib)
+![Github Actions](https://github.com/PufferAI/PufferLib/actions/workflows/install.yml/badge.svg)
 [![](https://dcbadge.vercel.app/api/server/spT4huaGYV?style=plastic)](https://discord.gg/spT4huaGYV)
 [![Twitter](https://img.shields.io/twitter/url/https/twitter.com/cloudposse.svg?style=social&label=Follow%20%40jsuarez5341)](https://twitter.com/jsuarez5341)
 

config/ocean/pysquared.ini

@@ -0,0 +1,21 @@
+[base]
+package = ocean
+env_name = puffer_pysquared
+policy_name = Policy
+rnn_name = Recurrent
+
+[env]
+num_envs = 1
+
+[train]
+total_timesteps = 40_000_000
+checkpoint_interval = 50
+num_envs = 12288
+num_workers = 12
+env_batch_size = 4096
+batch_size = 131072
+update_epochs = 1
+minibatch_size = 8192
+learning_rate = 0.0017
+anneal_lr = False
+device = cuda

config/ocean/trash_pickup.ini

@@ -0,0 +1,64 @@
+[base]
+package = ocean
+env_name = trash_pickup puffer_trash_pickup 
+policy_name = TrashPickup
+rnn_name = Recurrent
+
+[env]
+num_envs = 1024  # Recommended: 4096 (recommended start value) / num_agents 
+grid_size = 10
+num_agents = 4
+num_trash = 20
+num_bins = 1
+max_steps = 150
+report_interval = 32
+agent_sight_range = 5 # only used with 2D local crop obs space
+
+[train]
+total_timesteps = 100_000_000
+checkpoint_interval = 200
+num_envs = 2
+num_workers = 2
+env_batch_size = 1
+batch_size = 131072
+update_epochs = 1
+minibatch_size = 16384
+bptt_horizon = 8
+anneal_lr = False
+device = cuda
+learning_rate=0.001
+gamma = 0.95
+gae_lambda = 0.85
+vf_ceof = 0.4
+clip_coef = 0.1
+vf_clip_coef = 0.1
+ent_coef = 0.01
+
+[sweep.metric]
+goal = maximize
+name = environment/episode_return
+
+[sweep.parameters.train.parameters.learning_rate]
+distribution = log_uniform_values
+min = 0.000001
+max = 0.01
+
+[sweep.parameters.train.parameters.gamma]
+distribution = uniform
+min = 0
+max = 1
+
+[sweep.parameters.train.parameters.gae_lambda]
+distribution = uniform
+min = 0
+max = 1
+
+[sweep.parameters.train.parameters.update_epochs]
+distribution = int_uniform
+min = 1
+max = 4
+
+[sweep.parameters.train.parameters.ent_coef]
+distribution = log_uniform_values
+min = 1e-5
+max = 1e-1