[RLlib] Add DQN SoftQ learning test case. (#12712)

rllib/BUILD

@@ -172,6 +172,7 @@ py_test(
     data = [
         "tuned_examples/dqn/cartpole-simpleq.yaml",
         "tuned_examples/dqn/cartpole-dqn.yaml",
+        "tuned_examples/dqn/cartpole-dqn-softq.yaml",
         "tuned_examples/dqn/cartpole-dqn-param-noise.yaml",
     ],
     args = ["--yaml-dir=tuned_examples/dqn"]
@@ -186,6 +187,7 @@ py_test(
     data = [
         "tuned_examples/dqn/cartpole-simpleq.yaml",
         "tuned_examples/dqn/cartpole-dqn.yaml",
+        "tuned_examples/dqn/cartpole-dqn-softq.yaml",
         "tuned_examples/dqn/cartpole-dqn-param-noise.yaml",
     ],
     args = ["--yaml-dir=tuned_examples/dqn", "--torch"]

rllib/tuned_examples/dqn/cartpole-dqn-softq.yaml

@@ -0,0 +1,16 @@
+cartpole-dqn:
+    env: CartPole-v0
+    run: DQN
+    stop:
+        episode_reward_mean: 150
+        timesteps_total: 100000
+    config:
+        # Works for both torch and tf.
+        framework: tf
+        model:
+            fcnet_hiddens: [64]
+            fcnet_activation: linear
+        n_step: 3
+        exploration_config:
+            type: SoftQ
+            temperature: 0.5