init add of wandb ai writer

ml-agents/mlagents/plugins/stats_writer.py

@@ -13,7 +13,7 @@
 from mlagents_envs import logging_util
 from mlagents.plugins import ML_AGENTS_STATS_WRITER
 from mlagents.trainers.settings import RunOptions
-from mlagents.trainers.stats import TensorboardWriter, GaugeWriter, ConsoleWriter
+from mlagents.trainers.stats import TensorboardWriter, GaugeWriter, ConsoleWriter, WandbWriter
 
 
 logger = logging_util.get_logger(__name__)
@@ -25,6 +25,7 @@ def get_default_stats_writers(run_options: RunOptions) -> List[StatsWriter]:
     * A TensorboardWriter to write information to TensorBoard
     * A GaugeWriter to record our internal stats
     * A ConsoleWriter to output to stdout.
+    * A Wandb.AI Writer
     """
     checkpoint_settings = run_options.checkpoint_settings
     return [
@@ -35,6 +36,7 @@ def get_default_stats_writers(run_options: RunOptions) -> List[StatsWriter]:
         ),
         GaugeWriter(),
         ConsoleWriter(),
+        WandbWriter()
     ]
 
 

ml-agents/mlagents/trainers/stats.py

@@ -6,6 +6,7 @@
 import os
 import time
 from threading import RLock
+import wandb
 
 from mlagents_envs.side_channel.stats_side_channel import StatsAggregationMethod
 
@@ -286,6 +287,29 @@ def add_property(
                 self.summary_writers[category].flush()
 
 
+class WandbWriter(StatsWriter):
+    def __init__(
+        self,
+        config: dict
+    ):
+        """
+        A Weights and Biases Wrapper that will add stats to your wandb.ai board.
+        """
+        wandb.init(reinit=True,
+                   config=config)
+
+    def write_stats(
+        self,
+        category : str,
+        values   : dict,
+        step     : int
+    ) -> None:
+        """
+        Write some stats for a given category and step
+        """
+        wandb.log({category : values}, step=step)
+
+
 class StatsReporter:
     writers: List[StatsWriter] = []
     stats_dict: Dict[str, Dict[str, List]] = defaultdict(lambda: defaultdict(list))

ml-agents/mlagents/trainers/tests/test_stats.py

@@ -12,6 +12,7 @@
     StatsSummary,
     GaugeWriter,
     ConsoleWriter,
+    WandbWriter,
     StatsPropertyType,
     StatsAggregationMethod,
 )
@@ -248,3 +249,20 @@ def test_selfplay_console_writer(self):
         self.assertIn(
             "Mean Reward: 1.000. Std of Reward: 0.000. Training.", cm.output[0]
         )
+
+
+class WandbWriterTest(unittest.TestCase):
+    def test_wandb_full(self):
+        category = "GeneralStuff"
+        config   = {"caller" : "ml-agents"}
+        wandb_writer = WandbWriter(config=config)
+        wandb_writer.write_stats(
+            category = category,
+            values = {
+                "Environment/Cumulative Reward": -15,
+                "Is Training": True,
+                "Self-play/ELO": 1.0,
+            },
+            step = 10,
+        )
+