|
|
@@ -0,0 +1,77 @@
|
|
|
+import os
|
|
|
+from time import time
|
|
|
+
|
|
|
+import matplotlib.pyplot as plt
|
|
|
+import numpy as np
|
|
|
+import tensorflow as tf
|
|
|
+from icecream import ic
|
|
|
+from tensorflow import keras
|
|
|
+
|
|
|
+
|
|
|
+# os.environ['CUDA_VISIBLE_DEVICES'] = '-1'
|
|
|
+
|
|
|
+global_random_seed = 3884
|
|
|
+np.random.seed(global_random_seed)
|
|
|
+
|
|
|
+
|
|
|
+def gene_test_data(num_points: int, low_val=-1, high_val=1):
|
|
|
+ return np.linspace(low_val, high_val, num_points)
|
|
|
+
|
|
|
+
|
|
|
+def get_network(input_dim, output_dim, width, depth):
|
|
|
+ # Random weight
|
|
|
+ initializer = keras.initializers.RandomUniform(
|
|
|
+ minval=-1., maxval=1., seed=global_random_seed)
|
|
|
+ # initializer = keras.initializers.Constant(1)
|
|
|
+
|
|
|
+ # different activation functions
|
|
|
+ activation_func = keras.activations.tanh
|
|
|
+ # activation_func = tf.math.sin
|
|
|
+
|
|
|
+ hidden_layers = [keras.layers.InputLayer(input_shape=[input_dim, ])]
|
|
|
+ hidden_layers += [keras.layers.Dense(width, activation=activation_func,
|
|
|
+ kernel_initializer=initializer, bias_initializer=initializer) for _ in range(depth-1)]
|
|
|
+ hidden_layers.append(keras.layers.Dense(
|
|
|
+ output_dim, kernel_initializer=initializer))
|
|
|
+ model = keras.Sequential(hidden_layers)
|
|
|
+ return model
|
|
|
+
|
|
|
+
|
|
|
+def predict(input):
|
|
|
+ result = []
|
|
|
+ for model in all_models:
|
|
|
+ temp = model.predict(input)
|
|
|
+ result.append(temp)
|
|
|
+ return result
|
|
|
+
|
|
|
+
|
|
|
+model_5 = get_network(1, 1, 5, 15)
|
|
|
+model_10 = get_network(1, 1, 10, 15)
|
|
|
+model_20 = get_network(1, 1, 20, 15)
|
|
|
+model_40 = get_network(1, 1, 40, 15)
|
|
|
+all_models = [model_5, model_10, model_20, model_40]
|
|
|
+
|
|
|
+if __name__ == '__main__':
|
|
|
+ st_time = time()
|
|
|
+ test_data = gene_test_data(100)
|
|
|
+
|
|
|
+ result = predict(test_data)
|
|
|
+ ic(np.shape(result))
|
|
|
+ print(time()-st_time)
|
|
|
+ # result = list(zip(*result))
|
|
|
+ ic(np.shape(result))
|
|
|
+ result_fft = []
|
|
|
+ for i in range(len(result)):
|
|
|
+ result_fft.append(list(np.fft.fft(result[i], axis=0)))
|
|
|
+
|
|
|
+ ic(np.shape(result_fft))
|
|
|
+
|
|
|
+ fig, ((p1, p2, p3, p4), (p5, p6, p7, p8)) = plt.subplots(2, 4)
|
|
|
+ pics = [p1, p2, p3, p4, p5, p6, p7, p8]
|
|
|
+ for i in range(len(result)):
|
|
|
+ pics[i].scatter(range(len(result[i])), list(result[i]), s=1.5, c='r')
|
|
|
+ pics[i+4].scatter(range(len(result_fft[i])),
|
|
|
+ list(result_fft[i]), s=0.5, c='b')
|
|
|
+ pics[i].set(adjustable='box', aspect=1.0/pics[i].get_data_ratio())
|
|
|
+ pics[i+4].set(adjustable='box', aspect=1.0/pics[i+4].get_data_ratio())
|
|
|
+ plt.show()
|
