diff --git a/RandomBird/FlappyAgent.py b/RandomBird/FlappyAgent.py
new file mode 100644
index 0000000..d93b12a
--- /dev/null
+++ b/RandomBird/FlappyAgent.py
@@ -0,0 +1,154 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Jan 24 14:41:52 2018
+
+@author: Arnaud
+"""
+from ple.games.flappybird import FlappyBird
+from ple import PLE
+import numpy as np
+import pickle
+
+def epsilon_greedy(Q, y, dist,veloc, epsilon):
+    a = np.argmax(Q[y,dist,veloc,:])
+    if (a == 0) : 
+        a=None
+    else:
+        a=119
+    if(np.random.rand()<=epsilon): # random action
+        aa = np.random.randint(0,2)
+        if (aa == 0) : 
+            a=None
+        else:
+            a=119
+            
+    return a
+
+
+
+# Stockage des 9 dernières paires états-actions en cas de défaite
+def updateLast9(last9,y,dist,veloc,ia):
+    for i in range(0,9):
+        last9[i]=last9[i+1]
+    last9[9][0] = y
+    last9[9][1] = dist
+    last9[9][2] = veloc
+    last9[9][3] = ia
+
+nb_games = 1000000
+sizey = 15 #-300 à 300 #Différence entre le haut de next_pipe et l'oiseau 
+sizex = 10 # 0 à 283 #distance entre l'oiseau et next_pipe
+sizev = 9 #vitesse discretisée
+
+
+#Q-learning, matrice état + actions
+#Qql = np.zeros((sizey,sizex,sizev,2))
+epsilon = 0 #L'aléatoire n'est pas nécessaire, au contraire il réduit les performances car en pénalisant 9 pairs états-actions on risque de pénaliser N pairs intéressantes pour 9-N pairs aléatoires qui ont mené l'oiseau à sa perte
+
+r=0  #initialisation de la récompense
+
+
+#Parametres du modèle
+alpha = 0.05# Des valeurs de alpha de l'ordre de 0.4 étaient intéressante pour gagner des performances rapidement mais menaient systématiquement à du sur-apprentissage et à des pertes de performances (ex : montée à 80 de moyenne puis chute à 15 puis stagnation à 15)
+gamma = 0.95
+
+#Initialisation des états
+y = 0
+dist = 0
+veloc = 0
+ia = 0
+cumulated = np.zeros((nb_games))+5
+#
+## Read from file
+f_myfile = open('Q_functionArnaud.pickle', 'rb')
+Qql = pickle.load(f_myfile)  
+f_myfile.close()
+
+
+def FlappyPolicy(state, screen):
+
+    y=int((state['player_y']-state['next_pipe_bottom_y']+300)/40)
+    dist = int(state['next_pipe_dist_to_player']/40)
+    veloc = int((state['player_vel']+16)/3)
+    a = np.argmax(Qql[y,dist,veloc,:])
+   
+    if (a == 0) : 
+        a=None
+    else:
+        a=119 
+
+    return a
+##ENTRAINEMENT
+#train = True pour lancer l'entrainement
+train = False
+if train:
+    game = FlappyBird()
+    p = PLE(game,fps=30, frame_skip=1, num_steps=1, force_fps=True, display_screen=True)
+    p.init()
+
+    for i in range(0,nb_games):
+        p.reset_game()
+        state = game.getGameState()
+        screen = p.getScreenRGB()
+        y=int((state['player_y']-state['next_pipe_bottom_y']+300)/40)
+        dist = int(state['next_pipe_dist_to_player']/40)
+        veloc = int((state['player_vel']+16)/3)
+        last9 = np.zeros((10,4))
+        while(not p.game_over()):
+            
+        
+            if((i+1)%100==0):
+                epsilon = epsilon/2
+            if ((i+1)>300):
+                epsilon = 0
+            
+            a = epsilon_greedy(Qql,y,dist,veloc,epsilon)
+            if (a == None):
+                ia = 0
+            else:
+                ia = 1
+            
+            action=a ### Your job is to define this function.
+            rewGame = p.act(action)
+            state = game.getGameState()
+            screen = p.getScreenRGB()
+            ##Update de l'etat
+            yprec = y
+            distprec = dist
+            velocprec = veloc
+            y=int((state['player_y']-state['next_pipe_bottom_y']+300)/40)
+            dist = int(state['next_pipe_dist_to_player']/40)
+            veloc = int((state['player_vel']+16)/3)
+            
+            
+            r=1
+            updateLast9(last9,y,dist,veloc,ia)
+            Qql[yprec][distprec][velocprec][ia] = Qql[yprec][distprec][velocprec][ia] + alpha * (r+gamma*np.max(Qql[y][dist][veloc][:])-Qql[yprec][distprec][velocprec][ia])
+          
+           
+            cumulated[i] = cumulated[i] + rewGame
+            
+        for l in range(0,8):
+            r=-1000
+            yprec = int(last9[l][0])
+            y = int(last9[l+1][0])
+            distprec = int(last9[l][1])
+            dist = int(last9[l+1][1])
+            velocprec = int(last9[l][2])
+            veloc = int(last9[l+1][2])
+            ia = int(last9[l+1][3])
+            Qql[yprec][distprec][velocprec][ia] = Qql[yprec][distprec][velocprec][ia] + alpha * (r+gamma*np.max(Qql[y][dist][veloc][:])-Qql[yprec][distprec][velocprec][ia])
+        
+        if (i<100):
+            print("i = ",i," - cumulated[i] = ",cumulated[i]," - mean = ",np.mean(cumulated[0:i]))
+        else:
+            print("i = ",i," - cumulated[i] = ",cumulated[i]," - mean[-1000] = ",np.mean(cumulated[i-100:i]))
+            
+average_score = np.mean(cumulated[20000:24055])
+max_score = np.max(cumulated)
+#
+#print("saving model")
+#f_myfile = open('Q_functionArnaud.pickle', 'wb')
+#pickle.dump(Qql, f_myfile)
+#f_myfile.close()
+
diff --git a/RandomBird/Q_functionArnaud.pickle b/RandomBird/Q_functionArnaud.pickle
new file mode 100644
index 0000000..7d4408d
Binary files /dev/null and b/RandomBird/Q_functionArnaud.pickle differ
diff --git a/RandomBird/__pycache__/FlappyAgent.cpython-36.pyc b/RandomBird/__pycache__/FlappyAgent.cpython-36.pyc
new file mode 100644
index 0000000..61dd5e2
Binary files /dev/null and b/RandomBird/__pycache__/FlappyAgent.cpython-36.pyc differ
diff --git a/RandomBird/run.py b/RandomBird/run.py
new file mode 100644
index 0000000..49f21b8
--- /dev/null
+++ b/RandomBird/run.py
@@ -0,0 +1,29 @@
+# You're not allowed to change this file
+from ple.games.flappybird import FlappyBird
+from ple import PLE
+import numpy as np
+from FlappyAgent import FlappyPolicy
+
+game = FlappyBird(graphics="fixed") # use "fancy" for full background, random bird color and random pipe color, use "fixed" (default) for black background and constant bird and pipe colors.
+p = PLE(game, fps=30, frame_skip=1, num_steps=1, force_fps=True, display_screen=True)
+# Note: if you want to see you agent act in real time, set force_fps to False. But don't use this setting for learning, just for display purposes.
+
+p.init()
+reward = 0.0
+
+nb_games = 100
+cumulated = np.zeros((nb_games))
+
+for i in range(nb_games):
+    p.reset_game()
+    
+    while(not p.game_over()):
+        state = game.getGameState()
+        screen = p.getScreenRGB()
+        action=FlappyPolicy(state, screen) ### Your job is to define this function.
+        
+        reward = p.act(action)
+        cumulated[i] = cumulated[i] + reward
+
+average_score = np.mean(cumulated)
+max_score = np.max(cumulated)
diff --git a/RandomBird/sanstitre1.py b/RandomBird/sanstitre1.py
new file mode 100644
index 0000000..898e5d0
--- /dev/null
+++ b/RandomBird/sanstitre1.py
@@ -0,0 +1,11 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Sun Mar 11 23:04:31 2018
+
+@author: Arnaud
+"""
+import pickle
+f_myfile = open('Q_function_5000_ite.pickle', 'rb')
+Qql = pickle.load(f_myfile)  
+f_myfile.close()
+
diff --git a/THOMAS/FlappyAgent.py b/THOMAS/FlappyAgent.py
new file mode 100644
index 0000000..d93b12a
--- /dev/null
+++ b/THOMAS/FlappyAgent.py
@@ -0,0 +1,154 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Jan 24 14:41:52 2018
+
+@author: Arnaud
+"""
+from ple.games.flappybird import FlappyBird
+from ple import PLE
+import numpy as np
+import pickle
+
+def epsilon_greedy(Q, y, dist,veloc, epsilon):
+    a = np.argmax(Q[y,dist,veloc,:])
+    if (a == 0) : 
+        a=None
+    else:
+        a=119
+    if(np.random.rand()<=epsilon): # random action
+        aa = np.random.randint(0,2)
+        if (aa == 0) : 
+            a=None
+        else:
+            a=119
+            
+    return a
+
+
+
+# Stockage des 9 dernières paires états-actions en cas de défaite
+def updateLast9(last9,y,dist,veloc,ia):
+    for i in range(0,9):
+        last9[i]=last9[i+1]
+    last9[9][0] = y
+    last9[9][1] = dist
+    last9[9][2] = veloc
+    last9[9][3] = ia
+
+nb_games = 1000000
+sizey = 15 #-300 à 300 #Différence entre le haut de next_pipe et l'oiseau 
+sizex = 10 # 0 à 283 #distance entre l'oiseau et next_pipe
+sizev = 9 #vitesse discretisée
+
+
+#Q-learning, matrice état + actions
+#Qql = np.zeros((sizey,sizex,sizev,2))
+epsilon = 0 #L'aléatoire n'est pas nécessaire, au contraire il réduit les performances car en pénalisant 9 pairs états-actions on risque de pénaliser N pairs intéressantes pour 9-N pairs aléatoires qui ont mené l'oiseau à sa perte
+
+r=0  #initialisation de la récompense
+
+
+#Parametres du modèle
+alpha = 0.05# Des valeurs de alpha de l'ordre de 0.4 étaient intéressante pour gagner des performances rapidement mais menaient systématiquement à du sur-apprentissage et à des pertes de performances (ex : montée à 80 de moyenne puis chute à 15 puis stagnation à 15)
+gamma = 0.95
+
+#Initialisation des états
+y = 0
+dist = 0
+veloc = 0
+ia = 0
+cumulated = np.zeros((nb_games))+5
+#
+## Read from file
+f_myfile = open('Q_functionArnaud.pickle', 'rb')
+Qql = pickle.load(f_myfile)  
+f_myfile.close()
+
+
+def FlappyPolicy(state, screen):
+
+    y=int((state['player_y']-state['next_pipe_bottom_y']+300)/40)
+    dist = int(state['next_pipe_dist_to_player']/40)
+    veloc = int((state['player_vel']+16)/3)
+    a = np.argmax(Qql[y,dist,veloc,:])
+   
+    if (a == 0) : 
+        a=None
+    else:
+        a=119 
+
+    return a
+##ENTRAINEMENT
+#train = True pour lancer l'entrainement
+train = False
+if train:
+    game = FlappyBird()
+    p = PLE(game,fps=30, frame_skip=1, num_steps=1, force_fps=True, display_screen=True)
+    p.init()
+
+    for i in range(0,nb_games):
+        p.reset_game()
+        state = game.getGameState()
+        screen = p.getScreenRGB()
+        y=int((state['player_y']-state['next_pipe_bottom_y']+300)/40)
+        dist = int(state['next_pipe_dist_to_player']/40)
+        veloc = int((state['player_vel']+16)/3)
+        last9 = np.zeros((10,4))
+        while(not p.game_over()):
+            
+        
+            if((i+1)%100==0):
+                epsilon = epsilon/2
+            if ((i+1)>300):
+                epsilon = 0
+            
+            a = epsilon_greedy(Qql,y,dist,veloc,epsilon)
+            if (a == None):
+                ia = 0
+            else:
+                ia = 1
+            
+            action=a ### Your job is to define this function.
+            rewGame = p.act(action)
+            state = game.getGameState()
+            screen = p.getScreenRGB()
+            ##Update de l'etat
+            yprec = y
+            distprec = dist
+            velocprec = veloc
+            y=int((state['player_y']-state['next_pipe_bottom_y']+300)/40)
+            dist = int(state['next_pipe_dist_to_player']/40)
+            veloc = int((state['player_vel']+16)/3)
+            
+            
+            r=1
+            updateLast9(last9,y,dist,veloc,ia)
+            Qql[yprec][distprec][velocprec][ia] = Qql[yprec][distprec][velocprec][ia] + alpha * (r+gamma*np.max(Qql[y][dist][veloc][:])-Qql[yprec][distprec][velocprec][ia])
+          
+           
+            cumulated[i] = cumulated[i] + rewGame
+            
+        for l in range(0,8):
+            r=-1000
+            yprec = int(last9[l][0])
+            y = int(last9[l+1][0])
+            distprec = int(last9[l][1])
+            dist = int(last9[l+1][1])
+            velocprec = int(last9[l][2])
+            veloc = int(last9[l+1][2])
+            ia = int(last9[l+1][3])
+            Qql[yprec][distprec][velocprec][ia] = Qql[yprec][distprec][velocprec][ia] + alpha * (r+gamma*np.max(Qql[y][dist][veloc][:])-Qql[yprec][distprec][velocprec][ia])
+        
+        if (i<100):
+            print("i = ",i," - cumulated[i] = ",cumulated[i]," - mean = ",np.mean(cumulated[0:i]))
+        else:
+            print("i = ",i," - cumulated[i] = ",cumulated[i]," - mean[-1000] = ",np.mean(cumulated[i-100:i]))
+            
+average_score = np.mean(cumulated[20000:24055])
+max_score = np.max(cumulated)
+#
+#print("saving model")
+#f_myfile = open('Q_functionArnaud.pickle', 'wb')
+#pickle.dump(Qql, f_myfile)
+#f_myfile.close()
+
diff --git a/THOMAS/Q_functionArnaud.pickle b/THOMAS/Q_functionArnaud.pickle
new file mode 100644
index 0000000..7d4408d
Binary files /dev/null and b/THOMAS/Q_functionArnaud.pickle differ
diff --git a/THOMAS/__pycache__/FlappyAgent.cpython-36.pyc b/THOMAS/__pycache__/FlappyAgent.cpython-36.pyc
new file mode 100644
index 0000000..61dd5e2
Binary files /dev/null and b/THOMAS/__pycache__/FlappyAgent.cpython-36.pyc differ
diff --git a/THOMAS/run.py b/THOMAS/run.py
new file mode 100644
index 0000000..49f21b8
--- /dev/null
+++ b/THOMAS/run.py
@@ -0,0 +1,29 @@
+# You're not allowed to change this file
+from ple.games.flappybird import FlappyBird
+from ple import PLE
+import numpy as np
+from FlappyAgent import FlappyPolicy
+
+game = FlappyBird(graphics="fixed") # use "fancy" for full background, random bird color and random pipe color, use "fixed" (default) for black background and constant bird and pipe colors.
+p = PLE(game, fps=30, frame_skip=1, num_steps=1, force_fps=True, display_screen=True)
+# Note: if you want to see you agent act in real time, set force_fps to False. But don't use this setting for learning, just for display purposes.
+
+p.init()
+reward = 0.0
+
+nb_games = 100
+cumulated = np.zeros((nb_games))
+
+for i in range(nb_games):
+    p.reset_game()
+    
+    while(not p.game_over()):
+        state = game.getGameState()
+        screen = p.getScreenRGB()
+        action=FlappyPolicy(state, screen) ### Your job is to define this function.
+        
+        reward = p.act(action)
+        cumulated[i] = cumulated[i] + reward
+
+average_score = np.mean(cumulated)
+max_score = np.max(cumulated)
diff --git a/THOMAS/sanstitre1.py b/THOMAS/sanstitre1.py
new file mode 100644
index 0000000..898e5d0
--- /dev/null
+++ b/THOMAS/sanstitre1.py
@@ -0,0 +1,11 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Sun Mar 11 23:04:31 2018
+
+@author: Arnaud
+"""
+import pickle
+f_myfile = open('Q_function_5000_ite.pickle', 'rb')
+Qql = pickle.load(f_myfile)  
+f_myfile.close()
+