basic affine model

service/examples/affine/model_basic.py

@@ -0,0 +1,60 @@
+#-----------------------------------------------------------------------------------
+# Basic exmple showing affine function f(x) = ax + b 
+# 
+#-----------------------------------------------------------------------------------
+
+import numpy as np
+import tensorflow as tf
+import matplotlib.pyplot as plt
+
+
+x = np.array([1, 5, 8, 9 ,10, 15,13, 3,-2],np.float32)
+y = np.array([-2,-5, -7, -12 ,-15, -5, -12,-10,-5],np.float32)
+
+
+BATCH_SIZE  = 4
+num_epochs = 1000
+
+dataset = tf.data.Dataset.from_tensor_slices(( x , y ))
+
+
+dataset = tf.data.Dataset.from_tensor_slices(( x , y ))
+dataset = dataset.shuffle(9).batch(BATCH_SIZE, drop_remainder=True).repeat(num_epochs)
+
+
+model = tf.keras.Sequential()
+model.add(tf.keras.layers.Dense(units=1, activation='linear',input_shape=(1,)))
+
+model.layers[0].set_weights([np.array([[7.3]],np.float32),np.array([5.5],np.float32)])
+
+model.compile(optimizer='sgd', loss='mse')
+
+model.fit(dataset, epochs=num_epochs, steps_per_epoch=tf.data.experimental.cardinality(dataset).numpy() // BATCH_SIZE -1, verbose=0)
+
+print(model.layers[0].get_weights())
+
+
+
+# Save model to SavedModel format
+tf.saved_model.save(model, "./affine/")
+
+yy= [model((np.array([[xi]],np.float32))).numpy()[0,0] for xi in x]
+
+plt.figure(1)
+
+plt.scatter(x, y, c='b', label='Donnees')
+
+plt.plot(x, yy, c='r', label='model')
+
+plt.show()
+
+#result example    y=  -0.38 x  - 5.9
+
+x_in = [0,1]
+y_out = model.predict(x_in)
+
+print(y_out)
+
+
+delta =(y_out[0] -  y_out[1]) / ( x_in[0] - x_in[1])
+print(delta)

service/examples/load_model/create_model.py

@@ -13,7 +13,7 @@ def example_1():
     i = tf.initializers.global_variables()
 
     # Write the model definition
-    with open('models/load_model.pb', 'wb') as f:
+    with open('models/simple_model.pb', 'wb') as f:
         f.write(tf.get_default_graph().as_graph_def().SerializeToString())
 
 

service/examples/load_model/main.cpp

@@ -5,7 +5,7 @@
 #include <iomanip>
 
 int main() {
-    Model model("../model.pb");
+    Model model("../simle_model.pb");
     model.init();
 
     Tensor input_a{model, "input_a"};