codecov 90%, improve 18%

Author: ghzha0

EduSim/Envs/MBS/Env.py

@@ -56,8 +56,10 @@ def step(self, learning_item_id: int, *args, **kwargs):
         return observation, reward, done, info
 
     def n_step(self, learning_path, *args, **kwargs):
+        group = []
         for learning_item_id in learning_path:
-            self.step(learning_item_id)
+            group.append(self.step(learning_item_id))
+        return group
 
     def end_episode(self, *args, **kwargs):
         observation, probabilities, reward = self.exam()

EduSim/Envs/MBS/utils.py

@@ -204,12 +204,34 @@ def generalized_power_law(
         the probability of recalling an item
 
 
+    Examples
+    --------
+    >>> student_ability = [0.5,0.6]
+    >>> additional_student_ability = [0.1,0.1]
+    >>> additional_student_ability = np.asarray(additional_student_ability)
+    >>> item_difficulty = [0.5,0.6]
+    >>> additional_item_difficulty = [0.1,0.1]
+    >>> additional_item_difficulty = np.asarray(additional_item_difficulty)
+    >>> elapsed_time = [9,9]
+    >>> decay_rate = 1.0
+    >>> correct_times= [10,10]
+    >>> attempts= [20,20]
+    >>> [x for x in generalized_power_law(student_ability, \
+                additional_student_ability, \
+                item_difficulty, \
+                additional_item_difficulty, \
+                elapsed_time, \
+                decay_rate, \
+                correct_times, \
+                attempts, \
+                None, \
+                None)]
+    [0.2, 0.2]
     """
     r = decay_rate
     a = as_array(student_ability)
     d = as_array(item_difficulty)
     t = as_array(elapsed_time)
-
     if window_correct_coefficients is None:
         m = 1 / (1 + np.exp(-(a - d)))
     else:
@@ -227,18 +249,6 @@ def generalized_power_law(
 
     f = np.exp(additional_student_ability - additional_item_difficulty)
 
-    # if t < 0:
-    #     print("t<0")
-    #     exit()
-
-    # if np.isnan((m * (1 + r * t)) ** (-f)):
-    #     print((m * (1 + r * t)) ** (-f))
-    #     print("(m * (1 + r * t)):",(m * (1 + r * t)))
-    #     print("-f:",-f)
-    #     print("m:",m)
-    #     print("r:",r)
-    #     print("t:",t)
-
     return (m * (1 + r * t)) ** (-f)
 
 

EduSim/Envs/meta/Learner.py

@@ -49,6 +49,19 @@ class MetaLearnerGroup(object):
     pass
 
 
+"""
+Example:
+>>> learners = [MetaLearner(i) for i in range(10)]
+>>> mflg = MetaFiniteLearnerGroup(learners)
+>>> mflg.__len__()
+10
+>>> isinstance(mflg.__getitem__[0],MetaLearner)
+True
+>>> isinstance(mflg.sample(),MetaLearner)
+True
+"""
+
+
 class MetaFiniteLearnerGroup(MetaLearnerGroup):
     def __init__(self, learners: list, seed=None, *args, **kwargs):
         self._learners = learners
@@ -64,6 +77,12 @@ def __len__(self):
         return len(self._learners)
 
 
+"""
+>>> MLG = MetaLearnerGroup()
+>>> MetaInfinityLearnerGroup(MLG)
+"""
+
+
 class MetaInfinityLearnerGroup(MetaLearnerGroup):
     def __next__(self):
         raise NotImplementedError

EduSim/SimOS/SimOS.py

@@ -27,8 +27,8 @@ def step(self):
     def tune(self, *args, **kwargs):
         raise NotImplementedError
 
-    def n_step(self, max_steps: int):
-        return []
+    def n_step(self):
+        raise NotImplementedError
 
 
 class RandomAgent(MetaAgent):

examples/MBS/EFC.ipynb

@@ -78,7 +78,7 @@
     "    env,\n",
     "    max_steps=200,\n",
     "    max_episode_num=100,\n",
-    "    level=\"summary\",\n",
+    "    level=\"summary\"\n",
     ")\n",
     "print(\"done\")\n"
    ],

tensorBoard/events.out.tfevents.1615532062.DESKTOP-PBIIA72

@@ -0,0 +1,2 @@
+# coding: utf-8
+# 2021/03/12 @ zhaoguanhao

tensorBoard/events.out.tfevents.1615532063.DESKTOP-PBIIA72

@@ -0,0 +1,33 @@
+# coding: utf-8
+# 2021/03/12 @ guanhao,zhao
+
+import pytest
+from EduSim import EFCEnv
+
+REWARD = ["likelihood", "log_likelihood"]
+
+
+@pytest.mark.parametrize("reward", REWARD)
+def test_api(reward):
+    env = EFCEnv(n_items=30, reward_func=reward, threshold=0.5, seed=None, n_steps=200)
+    assert set(env.parameters.keys()) == {'action_space'}
+
+
+@pytest.mark.parametrize("reward", REWARD)
+@pytest.mark.parametrize("n_step", [True, False])
+@pytest.mark.parametrize("board_dir", [None, "tensorBoard"])
+def test_env(reward, n_step, board_dir):
+    from EduSim.Envs.MBS import EFCEnv, MBSAgent, mbs_train_eval
+    env = EFCEnv(n_items=30, reward_func=reward, threshold=0.5, seed=None, n_steps=200)
+    learner = next(env.learners)
+    learner.state
+    agent = MBSAgent(env.action_space)
+    mbs_train_eval(
+        agent,
+        env,
+        max_steps=200,
+        max_episode_num=3,
+        level="summary",
+        n_step=n_step,
+        board_dir=board_dir
+    )

tensorBoard/events.out.tfevents.1615532140.DESKTOP-PBIIA72

@@ -0,0 +1,33 @@
+# coding: utf-8
+# 2021/03/12 @ guanhao,zhao
+
+import pytest
+from EduSim import GPLEnv, HLREnv, EFCEnv
+
+REWARD = ["likelihood", "log_likelihood"]
+
+
+@pytest.mark.parametrize("reward", REWARD)
+def test_api(reward):
+    env = GPLEnv(n_items=30, reward_func=reward, threshold=0.5, seed=None, n_steps=200)
+    env.begin_episode()
+    env.render(mode="log")
+    assert set(env.parameters.keys()) == {'action_space'}
+
+
+@pytest.mark.parametrize("reward", REWARD)
+@pytest.mark.parametrize("n_step", [True, False])
+def test_env(reward, n_step):
+    from EduSim.Envs.MBS import GPLEnv, MBSAgent, mbs_train_eval
+    env = GPLEnv(n_items=30, reward_func=reward, threshold=0.5, seed=None, n_steps=200)
+    learner = next(env.learners)
+    learner.state
+    agent = MBSAgent(env.action_space)
+    mbs_train_eval(
+        agent,
+        env,
+        max_steps=200,
+        max_episode_num=1,
+        level="summary",
+        n_step=n_step
+    )

tensorBoard/events.out.tfevents.1615532141.DESKTOP-PBIIA72

@@ -0,0 +1,31 @@
+# coding: utf-8
+# 2021/03/12 @ guanhao,zhao
+
+import pytest
+from EduSim import HLREnv
+
+REWARD = ["likelihood", "log_likelihood"]
+
+
+@pytest.mark.parametrize("reward", REWARD)
+def test_api(reward):
+    env = HLREnv(n_items=30, reward_func=reward, threshold=0.5, seed=None, n_steps=200)
+    assert set(env.parameters.keys()) == {'action_space'}
+
+
+@pytest.mark.parametrize("reward", REWARD)
+@pytest.mark.parametrize("n_step", [True, False])
+def test_env(reward, n_step):
+    from EduSim.Envs.MBS import HLREnv, MBSAgent, mbs_train_eval
+    env = HLREnv(n_items=30, reward_func=reward, threshold=0.5, seed=None, n_steps=200)
+    learner = next(env.learners)
+    learner.state
+    agent = MBSAgent(env.action_space)
+    mbs_train_eval(
+        agent,
+        env,
+        max_steps=200,
+        max_episode_num=3,
+        level="summary",
+        n_step=n_step
+    )