avancement sur les fonctions d'activation

code/fonctions_activations_classiques/fonction_activation.py

@@ -0,0 +1,22 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Nov 24 14:48:00 2021
+
+@author: virgi
+"""
+import tensorflow as tf
+
+
+
+def snake(x, alpha=1.0):
+    return (x + tf.sin(x)**2/alpha)
+
+
+def x_sin(x,alpha=1.0):
+    return (x + tf.sin(x)/alpha)
+
+def sin(x,alpha=1.0):
+    return(tf.sin(x)/alpha)
+
+
+

code/fonctions_activations_classiques/premier_reseau.py

@@ -4,7 +4,7 @@ Created on Wed Nov 10 10:30:04 2021
 
 @author: virgi
 """
-
+from Creation_donnee import *
 import numpy as np
 import math as ma
 import matplotlib.pyplot as plt
@@ -20,13 +20,13 @@ Xv,Yv=creation_sin(tmin,tmax,n,1,a=1,c=0)
 model = tf.keras.models.Sequential()
 
 model.add(tf.keras.Input(shape=(1,)))
-# une première couche avec deux neurones (couche cachée) avec la fonction sigmoid comme fonction d'activation
+
 model.add(tf.keras.layers.Dense(64, activation='relu'))
 model.add(tf.keras.layers.Dense(64, activation='relu'))
 model.add(tf.keras.layers.Dense(64, activation='relu'))
 model.add(tf.keras.layers.Dense(64, activation='relu'))
 model.add(tf.keras.layers.Dense(128, activation='relu'))
-# une deuxième couche avec un neurones (couche de sortie) avec la fonction sigmoid comme fonction d'activation
+
 
 model.add(tf.keras.layers.Dense(1))
 

code/fonctions_activations_classiques/sin.py

@@ -0,0 +1,77 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Nov 24 16:58:44 2021
+
+@author: virgi
+"""
+
+
+
+
+
+import tensorflow as tf
+import matplotlib.pyplot as plt
+from fonction_activation import *
+
+from Creation_donnee import *
+import numpy as np
+
+n=20
+#création de la base de donnéé
+X,Y=creation_sin(-15,-8,n,1,)
+X2,Y2=creation_sin(10,18,n,1,)
+X=np.concatenate([X,X2])
+Y=np.concatenate([Y,Y2])
+
+n=10000
+Xv,Yv=creation_sin(-20,20,n,1)
+
+
+
+
+model_sin=tf.keras.models.Sequential()
+
+model_sin.add(tf.keras.Input(shape=(1,)))
+
+model_sin.add(tf.keras.layers.Dense(4, activation=sin))
+model_sin.add(tf.keras.layers.Dense(4, activation=sin))
+model_sin.add(tf.keras.layers.Dense(4, activation=sin))
+model_sin.add(tf.keras.layers.Dense(4, activation=sin))
+model_sin.add(tf.keras.layers.Dense(4, activation=sin))
+
+
+model_sin.add(tf.keras.layers.Dense(1))
+
+opti=tf.keras.optimizers.Adam()
+
+model_sin.compile(opti, loss='mse', metrics=['accuracy'])
+
+
+model_sin.summary()
+
+model_sin.fit(X, Y, batch_size=1, epochs=10, shuffle='True',validation_data=(Xv, Yv))
+
+
+
+
+Y_predis_sin=model_sin.predict(X)
+Y_predis_validation_sin=model_sin.predict(Xv)
+
+
+
+
+plt.figure()
+plt.plot(X,Y,'x',label='donnée')
+plt.plot(Xv,Yv,label="validation")
+plt.plot(X,Y_predis_sin,'o',label='prediction sur les donné avec sin comme activation')
+plt.plot(Xv,Y_predis_validation_sin,label='prediction sur la validation avec sin comme activation')
+plt.legend()
+plt.show()
+
+
+"""
+Created on Wed Nov 24 16:53:37 2021
+
+@author: virgi
+"""
+

code/fonctions_activations_classiques/snake_vs_ReLU.py

@@ -0,0 +1,108 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Nov 24 15:44:12 2021
+
+@author: virgi
+"""
+
+import tensorflow as tf
+import matplotlib.pyplot as plt
+from fonction_activation import *
+
+from Creation_donnee import *
+import numpy as np
+
+n=20
+#création de la base de donnéé
+X,Y=creation_sin(-15,-8,n,1,)
+X2,Y2=creation_sin(10,18,n,1,)
+X=np.concatenate([X,X2])
+Y=np.concatenate([Y,Y2])
+
+n=10000
+Xv,Yv=creation_sin(-20,20,n,1)
+
+
+model_ReLU=tf.keras.models.Sequential()
+
+
+model_ReLU.add(tf.keras.Input(shape=(1,)))
+
+model_ReLU.add(tf.keras.layers.Dense(64, activation='relu'))
+model_ReLU.add(tf.keras.layers.Dense(64, activation='relu'))
+model_ReLU.add(tf.keras.layers.Dense(64, activation='relu'))
+model_ReLU.add(tf.keras.layers.Dense(64, activation='relu'))
+model_ReLU.add(tf.keras.layers.Dense(64, activation='relu'))
+
+
+model_ReLU.add(tf.keras.layers.Dense(1))
+
+# Choix de la méthode d'optimisation
+opti=tf.keras.optimizers.Adam()
+# Compilation du graphe et choix de la fonction de coût
+model_ReLU.compile(opti, loss='mse', metrics=['accuracy'])
+
+model_ReLU.summary()
+
+
+
+model_ReLU.fit(X, Y, batch_size=1, epochs=10, shuffle='True',validation_data=(Xv, Yv))
+
+
+
+model_snake=tf.keras.models.Sequential()
+
+model_snake.add(tf.keras.Input(shape=(1,)))
+
+model_snake.add(tf.keras.layers.Dense(64, activation=snake))
+model_snake.add(tf.keras.layers.Dense(64, activation=snake))
+model_snake.add(tf.keras.layers.Dense(64, activation=snake))
+model_snake.add(tf.keras.layers.Dense(64, activation=snake))
+model_snake.add(tf.keras.layers.Dense(64, activation=snake))
+
+
+model_snake.add(tf.keras.layers.Dense(1))
+
+opti=tf.keras.optimizers.Adam()
+
+model_snake.compile(opti, loss='mse', metrics=['accuracy'])
+
+
+model_snake.summary()
+
+model_snake.fit(X, Y, batch_size=1, epochs=10, shuffle='True',validation_data=(Xv, Yv))
+
+
+
+
+
+
+
+
+
+Y_predis_snake=model_snake.predict(X)
+Y_predis_validation_snake=model_snake.predict(Xv)
+
+Y_predis_ReLU=model_ReLU.predict(X)
+Y_predis_validation_ReLU=model_ReLU.predict(Xv)
+
+
+
+
+
+plt.figure()
+plt.plot(X,Y,'x',label='donnée')
+plt.plot(Xv,Yv,label="validation")
+plt.plot(X,Y_predis_ReLU,'o',label='prediction sur les donné')
+plt.plot(Xv,Y_predis_validation_ReLU,label='prediction sur la validation')
+plt.legend()
+plt.show()
+
+plt.figure()
+plt.plot(X,Y,'x',label='donnée')
+plt.plot(Xv,Yv,label="validation")
+plt.plot(X,Y_predis_snake,'o',label='prediction sur les donné avec snake comme activation')
+plt.plot(Xv,Y_predis_validation_snake,label='prediction sur la validation avec snake comme activation')
+plt.legend()
+plt.show()
+

code/fonctions_activations_classiques/swish.py

@@ -0,0 +1,75 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Nov 24 16:48:52 2021
+
+@author: virgi
+"""
+
+
+
+import tensorflow as tf
+import matplotlib.pyplot as plt
+from fonction_activation import *
+
+from Creation_donnee import *
+import numpy as np
+
+n=20
+#création de la base de donnéé
+X,Y=creation_sin(-15,-8,n,1,)
+X2,Y2=creation_sin(10,18,n,1,)
+X=np.concatenate([X,X2])
+Y=np.concatenate([Y,Y2])
+
+n=10000
+Xv,Yv=creation_sin(-20,20,n,1)
+
+
+
+
+model_swish=tf.keras.models.Sequential()
+
+model_swish.add(tf.keras.Input(shape=(1,)))
+
+model_swish.add(tf.keras.layers.Dense(64, activation='swish'))
+model_swish.add(tf.keras.layers.Dense(64, activation='swish'))
+model_swish.add(tf.keras.layers.Dense(64, activation='swish'))
+model_swish.add(tf.keras.layers.Dense(64, activation='swish'))
+model_swish.add(tf.keras.layers.Dense(64, activation='swish'))
+
+
+model_swish.add(tf.keras.layers.Dense(1))
+
+opti=tf.keras.optimizers.Adam()
+
+model_swish.compile(opti, loss='mse', metrics=['accuracy'])
+
+
+model_swish.summary()
+
+model_swish.fit(X, Y, batch_size=1, epochs=10, shuffle='True',validation_data=(Xv, Yv))
+
+
+
+
+Y_predis_swish=model_swish.predict(X)
+Y_predis_validation_swish=model_swish.predict(Xv)
+
+
+
+
+plt.figure()
+plt.plot(X,Y,'x',label='donnée')
+plt.plot(Xv,Yv,label="validation")
+plt.plot(X,Y_predis_swish,'o',label='prediction sur les donné avec swish comme activation')
+plt.plot(Xv,Y_predis_validation_swish,label='prediction sur la validation avec swish comme activation')
+plt.legend()
+plt.show()
+
+
+"""
+Created on Wed Nov 24 16:53:37 2021
+
+@author: virgi
+"""
+

code/fonctions_activations_classiques/tanh_vs_ReLU.py

@@ -0,0 +1,110 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Nov 24 16:48:52 2021
+
+@author: virgi
+"""
+
+
+
+import tensorflow as tf
+import matplotlib.pyplot as plt
+from fonction_activation import *
+
+from Creation_donnee import *
+import numpy as np
+
+n=20
+#création de la base de donnéé
+X,Y=creation_sin(-15,-8,n,1,)
+X2,Y2=creation_sin(10,18,n,1,)
+X=np.concatenate([X,X2])
+Y=np.concatenate([Y,Y2])
+
+n=10000
+Xv,Yv=creation_sin(-20,20,n,1)
+
+
+model_ReLU=tf.keras.models.Sequential()
+
+
+model_ReLU.add(tf.keras.Input(shape=(1,)))
+
+model_ReLU.add(tf.keras.layers.Dense(64, activation='relu'))
+model_ReLU.add(tf.keras.layers.Dense(64, activation='relu'))
+model_ReLU.add(tf.keras.layers.Dense(64, activation='relu'))
+model_ReLU.add(tf.keras.layers.Dense(64, activation='relu'))
+model_ReLU.add(tf.keras.layers.Dense(64, activation='relu'))
+
+
+model_ReLU.add(tf.keras.layers.Dense(1))
+
+# Choix de la méthode d'optimisation
+opti=tf.keras.optimizers.Adam()
+# Compilation du graphe et choix de la fonction de coût
+model_ReLU.compile(opti, loss='mse', metrics=['accuracy'])
+
+model_ReLU.summary()
+
+
+
+model_ReLU.fit(X, Y, batch_size=1, epochs=10, shuffle='True',validation_data=(Xv, Yv))
+
+
+
+model_tanh=tf.keras.models.Sequential()
+
+model_tanh.add(tf.keras.Input(shape=(1,)))
+
+model_tanh.add(tf.keras.layers.Dense(64, activation='tanh'))
+model_tanh.add(tf.keras.layers.Dense(64, activation='tanh'))
+model_tanh.add(tf.keras.layers.Dense(64, activation='tanh'))
+model_tanh.add(tf.keras.layers.Dense(64, activation='tanh'))
+model_tanh.add(tf.keras.layers.Dense(64, activation='tanh'))
+
+
+model_tanh.add(tf.keras.layers.Dense(1))
+
+opti=tf.keras.optimizers.Adam()
+
+model_tanh.compile(opti, loss='mse', metrics=['accuracy'])
+
+
+model_tanh.summary()
+
+model_tanh.fit(X, Y, batch_size=1, epochs=10, shuffle='True',validation_data=(Xv, Yv))
+
+
+
+
+
+
+
+
+
+Y_predis_tanh=model_tanh.predict(X)
+Y_predis_validation_tanh=model_tanh.predict(Xv)
+
+Y_predis_ReLU=model_ReLU.predict(X)
+Y_predis_validation_ReLU=model_ReLU.predict(Xv)
+
+
+
+
+
+plt.figure()
+plt.plot(X,Y,'x',label='donnée')
+plt.plot(Xv,Yv,label="validation")
+plt.plot(X,Y_predis_ReLU,'o',label='prediction sur les donné')
+plt.plot(Xv,Y_predis_validation_ReLU,label='prediction sur la validation')
+plt.legend()
+plt.show()
+
+plt.figure()
+plt.plot(X,Y,'x',label='donnée')
+plt.plot(Xv,Yv,label="validation")
+plt.plot(X,Y_predis_tanh,'o',label='prediction sur les donné avec tanh comme activation')
+plt.plot(Xv,Y_predis_validation_tanh,label='prediction sur la validation avec tanh comme activation')
+plt.legend()
+plt.show()
+

code/fonctions_activations_classiques/x_sin.py

@@ -0,0 +1,84 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Nov 24 17:02:16 2021
+
+@author: virgi
+"""
+
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Nov 24 16:58:44 2021
+
+@author: virgi
+"""
+
+
+
+
+
+import tensorflow as tf
+import matplotlib.pyplot as plt
+from fonction_activation import *
+
+from Creation_donnee import *
+import numpy as np
+
+n=20
+#création de la base de donnéé
+X,Y=creation_sin(-15,-8,n,1,)
+X2,Y2=creation_sin(10,18,n,1,)
+X=np.concatenate([X,X2])
+Y=np.concatenate([Y,Y2])
+
+n=10000
+Xv,Yv=creation_sin(-20,20,n,1)
+
+
+
+
+model_xsin=tf.keras.models.Sequential()
+
+model_xsin.add(tf.keras.Input(shape=(1,)))
+
+model_xsin.add(tf.keras.layers.Dense(64, activation=x_sin))
+model_xsin.add(tf.keras.layers.Dense(64, activation=x_sin))
+model_xsin.add(tf.keras.layers.Dense(64, activation=x_sin))
+model_xsin.add(tf.keras.layers.Dense(64, activation=x_sin))
+model_xsin.add(tf.keras.layers.Dense(64, activation=x_sin))
+
+
+model_xsin.add(tf.keras.layers.Dense(1))
+
+opti=tf.keras.optimizers.Adam()
+
+model_xsin.compile(opti, loss='mse', metrics=['accuracy'])
+
+
+model_xsin.summary()
+
+model_xsin.fit(X, Y, batch_size=1, epochs=10, shuffle='True',validation_data=(Xv, Yv))
+
+
+
+
+Y_predis_xsin=model_xsin.predict(X)
+Y_predis_validation_xsin=model_xsin.predict(Xv)
+
+
+
+
+plt.figure()
+plt.plot(X,Y,'x',label='donnée')
+plt.plot(Xv,Yv,label="validation")
+plt.plot(X,Y_predis_xsin,'o',label='prediction sur les donné avec x+sin comme activation')
+plt.plot(Xv,Y_predis_validation_xsin,label='prediction sur la validation avec x+sin comme activation')
+plt.legend()
+plt.show()
+
+
+"""
+Created on Wed Nov 24 16:53:37 2021
+
+@author: virgi
+"""
+