tf RNN example2

Author: Mofan Zhou

RL/example1/hunter_prey.py

@@ -0,0 +1,130 @@
+import tkinter as tk
+import numpy as np
+import pandas as pd
+
+
+class TableLUQ(object):
+    def __init__(self, actions, epsilon=0.1, alpha=0.2, gamma=0.9):
+        self._actions = actions
+        self._epsilon = epsilon
+        self._alpha = alpha
+        self._gamma = gamma
+
+        # state-actions table. columns include [state, a1, a2, a3...]
+        self._q_table = pd.DataFrame(columns=['state']+self._actions)
+        self._q_table.set_index('state', inplace=True)
+
+    def learn(self, s0, a0, r, s1):
+        q_s0_a0 = self._q_table.loc[s0, a0]
+        if s1 in self._q_table.index:
+            q_s1_a = self._q_table.loc[s1, :]
+            q_s1_a_max = q_s1_a.max()
+        else:
+            q_s1_a_max = 0
+        self._q_table.loc[s0, a0] = q_s0_a0 + self._alpha * (
+            r + self._gamma * q_s1_a_max - q_s0_a0
+        )
+
+    def choose_action(self, state):
+        if np.random.random() < self._epsilon:
+            action = np.random.choice(self._actions)
+        else:
+            if state not in self._q_table.index:
+                new_state_actions = pd.Series([0]*len(self._actions), index=self._actions, name=state)
+                self._q_table = self._q_table.append(new_state_actions)
+                action = np.random.choice(self._actions)
+            else:
+                state_actions_pair = self._q_table.loc[state, :]
+                shuffled_state_actions_pair = state_actions_pair.reindex(
+                    np.random.permutation(
+                        state_actions_pair.index))
+                action = shuffled_state_actions_pair.argmax()
+        return action
+
+    @property
+    def epsilon(self):
+        return self._epsilon
+
+    @property
+    def alpha(self):
+        return self._alpha
+
+    @property
+    def gamma(self):
+        return self._gamma
+
+
+def get_state(h_loc, p_loc):
+    state = str(
+        (
+            p_loc[0] - h_loc[0],
+            p_loc[1] - h_loc[1]
+         )
+    )
+    return state
+
+
+def get_reward(h_loc, p_loc):
+    distance = ((p_loc[0] - h_loc[0])**2 + (p_loc[1] - h_loc[1])**2)**(1/2)
+    reward = 1/distance
+    return reward
+
+
+def move(h_loc, action):
+    h_x, h_y = h_loc[0], h_loc[1]
+    if action == 'u':
+        h_y -= 1
+        h_y = max([h_y, 0])
+    elif action == 'd':
+        h_y += 1
+        h_y = min([h_y, 4])
+    elif action == 'l':
+        h_x -= 1
+        h_x = max([h_x, 0])
+    else:
+        h_x += 1
+        h_x = min([h_x, 4])
+    x_amount = (h_x - h_loc[0])*100
+    y_amount = (h_y - h_loc[1])*100
+    canvas.move(hunter_icon, x_amount, y_amount)
+    h_loc = [h_x, h_y]
+
+window = tk.Tk()
+window.geometry('500x500')
+canvas = tk.Canvas(window, height=500, width=500)
+canvas.pack()
+
+hunter_loc = [0, 0]     # can move
+prey_loc = [4, 4]       # fixed
+
+hunter_icon = canvas.create_rectangle(
+    (
+        hunter_loc[0]*100,
+        hunter_loc[1]*100,
+        hunter_loc[0]*100+100,
+        hunter_loc[1]*100+100),
+    fill='black'
+)
+prey_icon = canvas.create_oval(
+    (
+        prey_loc[0]*100,
+        prey_loc[1]*100,
+        prey_loc[0]*100+100,
+        prey_loc[1]*100+100),
+    fill='red'
+)
+
+move_up = [+1, 0]
+move_down = [-1, 0]
+move_left = [0, -1]
+move_right = [0, +1]
+
+hunter_actions = {
+    'u': move_up,
+    'd': move_down,
+    'l': move_left,
+    'r': move_right,
+}
+
+
+window.mainloop()