Added test code for integrating Tensorforce

 * Need TF v1.7.0
 * lefnire/tforce_btc_trader#33

helper/baselines/__init__.py

@@ -1,4 +1,4 @@
 from .RandomAgent import RandomAgent
 from .FixedActionAgent import FixedActionAgent
 from .keras import *
-
+from .tensorforce import *

helper/baselines/tensorforce/TensorforcePPOAgent.py

@@ -0,0 +1,32 @@
+from tensorforce.agents import PPOAgent
+from helper.wrappers import ClientToEnv, DictToListFull, JSONable
+
+class TensorforcePPOAgent:
+    def __init__(self, env):
+        # Create a Proximal Policy Optimization agent
+        self.agent = PPOAgent(
+            states=dict(type='float', shape=(347,)),
+            actions=dict(type='float', shape=(19,)),
+            network=[
+                dict(type='dense', size=64),
+                dict(type='dense', size=64)
+            ],
+            batching_capacity=1000,
+            step_optimizer=dict(
+                type='adam',
+                learning_rate=1e-4
+            )
+        )
+
+        self.env = DictToListFull(env)
+
+    def test(self, env):
+        # Poll new state from client
+        state = self.env.reset(project=False)
+        import numpy as np
+        # Get prediction from agent, execute
+        action = self.agent.act(state)
+        obs, rew, done, info = self.env.step(action, project=False)
+
+        # Add experience, agent automatically updates model according to batch size
+        self.agent.observe(reward=rew, terminal=done)

helper/baselines/tensorforce/__init__.py

@@ -0,0 +1 @@
+from .TensorforcePPOAgent import TensorforcePPOAgent