feat(mathy_alpha_sm): add pretrained model with simplified architecture

 - remove time-dimension LSTM initial state handling. I had to tile the LSTM state to get this to work and it felt really hacky. It also expanded the size of the data by quite a bit with duplication. Removing this and going with a single LSTM works just as well or better.
 - adds a dense layer to the value head to fix layernorm issue with 1 unit output.

Author: justindujardin

libraries/mathy_alpha_sm/mathy_alpha_sm/model.config.json

@@ -6,47 +6,52 @@
   "author": "Justin DuJardin",
   "email": "justin@dujardinconsulting.com",
   "url": "https://mathy.ai",
-  "mathy_version": ">=0.5.1,<1.0.0",
-  "units": 512,
-  "embedding_units": 512,
-  "lstm_units": 128,
-  "topics": [
+  "mathy_version":">=0.6.0,<1.0.0",
+  "prediction_window_size":7,
+  "use_lstm":true,
+  "units":512,
+  "embedding_units":512,
+  "lstm_units":128,
+  "topics":[
     "poly",
-    "binomial",
-    "complex"
+    "complex",
+    "binomial"
   ],
-  "difficulty": "easy",
-  "init_model_from": null,
-  "train": false,
-  "verbose": false,
-  "lr": 0.0003,
-  "max_eps": 15000,
-  "summary_interval": 100,
-  "gamma": 0.99,
-  "num_workers": 4,
-  "td_lambda": 0.2,
-  "print_training": false,
-  "print_mode": "terminal",
-  "use_term_order": false,
-  "use_tree_complexity": false,
-  "use_grouping_control": true,
-  "clip_grouping_control": false,
-  "use_env_features": false,
-  "use_node_values": true,
-  "update_gradients_every": 64,
-  "num_thinking_steps_begin": 3,
-  "mcts_sims": 10,
-  "worker_wait": 0.1,
-  "profile": false,
-  "action_strategy": "a3c",
-  "main_worker_use_epsilon": false,
-  "e_greedy_min": 0.01,
-  "e_greedy_max": 0.1,
-  "entropy_loss_scaling": 0.05,
-  "normalize_entropy_loss": true,
-  "aux_tasks_weight_scale": 0.1,
-  "teacher_start_evaluations_at_episode": 50,
-  "teacher_evaluation_steps": 20,
-  "teacher_promote_wins": 0.95,
-  "teacher_demote_wins": 0.6
-}
+  "difficulty":null,
+  "model_name":"model",
+  "init_model_from":null,
+  "train":false,
+  "verbose":true,
+  "lr":0.001,
+  "max_eps":5000,
+  "summary_interval":100,
+  "gamma":0.99,
+  "num_workers":4,
+  "td_lambda":0.2,
+  "print_training":false,
+  "print_model_call_times":false,
+  "print_mode":"terminal",
+  "normalization_style":"layer",
+  "use_term_order":false,
+  "use_tree_complexity":false,
+  "use_grouping_control":false,
+  "clip_grouping_control":true,
+  "use_env_features":true,
+  "use_node_values":true,
+  "update_gradients_every":6,
+  "num_thinking_steps_begin":3,
+  "mcts_sims":10,
+  "worker_wait":0.01,
+  "profile":false,
+  "action_strategy":"a3c",
+  "main_worker_use_epsilon":false,
+  "e_greedy_min":0.01,
+  "e_greedy_max":0.3,
+  "entropy_loss_scaling":0.05,
+  "normalize_entropy_loss":true,
+  "aux_tasks_weight_scale":0.1,
+  "teacher_start_evaluations_at_episode":50,
+  "teacher_evaluation_steps":5,
+  "teacher_promote_wins":0.95,
+  "teacher_demote_wins":0.6
+}