initial commit

generic.py

@@ -0,0 +1,51 @@
+import tensorflow as tf
+import numpy as np
+from matplotlib import pyplot as plt
+import itertools as it
+
+from sklearn.metrics import r2_score
+
+from math import sin
+
+def f(x, y): return (1 + sin(x**2 + 5*y)) / 2
+
+def train_generic(model, marker):
+    model.compile(optimizer = "adam", loss = "mse")
+
+    X = np.linspace(0, 10, 300)
+    Y = np.linspace(0, 10, 300)
+
+    ins = np.array(list(it.product(X, Y)))
+    ous = np.array(list(f(*i) for i in ins))
+
+    result = model.fit(ins, ous, epochs = 200, batch_size = 2048)
+
+    model.save_weights(f"save-{marker}.weights.h5")
+
+def verify_generic(model, marker):
+    model.compile(optimizer = "adam", loss = "mse")
+    model.load_weights(f"save-{marker}.weights.h5")
+
+    X = np.linspace(0, 10, 50)
+    Y = np.linspace(0, 10, 50)
+
+    ins = np.array(list(it.product(X, Y)))
+    ous = np.array(list(f(*i) for i in ins))
+
+    preds = model.predict(ins)
+
+    print(f"Model {marker} has {r2_score(ous, preds)} r2 score")
+
+    preds_flat = [i[0] for i in preds]
+
+    px = np.array([X] * len(X))
+    py = np.array([[x] * (len(X)) for x in X])
+    pz1 = np.array([ous[i*len(X):(i+1)*len(X)] for i, _ in enumerate(X)])
+    pz2 = np.array([preds_flat[i*len(X):(i+1)*len(X)] for i, _ in enumerate(X)])
+    
+    #print([ous[i*len(X):(i+1)*len(X)] for i, _ in enumerate(X)])
+
+    p = plt.figure().add_subplot(projection='3d')
+    p.plot_surface(px, py, pz1, edgecolor = "lime", alpha = 0.1)
+    p.plot_surface(px, py, pz2, edgecolor = "blue", alpha = 0.3)
+    plt.show()

nn1-normal.py

@@ -0,0 +1,13 @@
+import tensorflow as tf
+import tensorflow.keras.layers as l
+
+import generic as g
+
+m = tf.keras.models.Sequential([
+    l.Input(shape = (2,)),
+    l.Dense(5, activation='relu'),
+    l.Dense(1)
+])
+
+#g.train_generic(m, "ff")
+g.verify_generic(m, "ff")

nn10-elman.py

@@ -0,0 +1,23 @@
+import tensorflow as tf
+import tensorflow.keras.layers as l
+
+import generic as g
+
+m = tf.keras.models.Sequential([
+    l.Input(shape = (2, 1)),
+    l.SimpleRNN(100, activation = "relu", return_sequences = True),
+    l.SimpleRNN(100, activation = "relu", return_sequences = True),
+    l.SimpleRNN(100, activation = "relu"),
+    l.Dense(1)
+])
+#i = l.Input(shape = (2,))
+#h1 = l.SimpleRNN(10, activation = "relu", input_shape = (2, 1))(i)
+#o = l.Dense(1)(h1)
+
+
+#m = tf.keras.models.Model(inputs = i, outputs = o)
+
+m.summary()
+
+#g.train_generic(m, "r3")
+g.verify_generic(m, "r3")

nn2-normal.py

@@ -0,0 +1,13 @@
+import tensorflow as tf
+import tensorflow.keras.layers as l
+
+import generic as g
+
+m = tf.keras.models.Sequential([
+    l.Input(shape = (2,)),
+    l.Dense(1000, activation = "relu"),
+    l.Dense(1)
+])
+
+#g.train_generic(m, "ff2")
+g.verify_generic(m, "ff2")

nn3-normal.py

@@ -0,0 +1,22 @@
+import tensorflow as tf
+import tensorflow.keras.layers as l
+
+import generic as g
+
+m = tf.keras.models.Sequential([
+    l.Input(shape = (2,)),
+    l.Dense(100, activation = "relu"),
+    l.Dense(100, activation = "relu"),
+    l.Dense(100, activation = "relu"),
+    l.Dense(100, activation = "relu"),
+    l.Dense(100, activation = "relu"),
+    l.Dense(100, activation = "relu"),
+    l.Dense(100, activation = "relu"),
+    l.Dense(100, activation = "relu"),
+    l.Dense(100, activation = "relu"),
+    l.Dense(100, activation = "relu"),
+    l.Dense(1)
+])
+
+#g.train_generic(m, "ff3")
+g.verify_generic(m, "ff3")

nn4-normal.py

@@ -0,0 +1,17 @@
+import tensorflow as tf
+import tensorflow.keras.layers as l
+
+import generic as g
+
+m = tf.keras.models.Sequential([
+    l.Input(shape = (2,)),
+    l.Dense(500, activation = "relu"),
+    l.Dense(500, activation = "relu"),
+    l.Dense(500, activation = "relu"),
+    l.Dense(500, activation = "relu"),
+    l.Dense(500, activation = "relu"),
+    l.Dense(1)
+])
+
+#g.train_generic(m, "ff4")
+g.verify_generic(m, "ff4")

nn5-cascade.py

@@ -0,0 +1,19 @@
+import tensorflow as tf
+import tensorflow.keras.layers as l
+
+import generic as g
+
+
+i = l.Input(shape = (2,))
+h1 = l.Dense(20, activation='relu')(i)
+
+y = l.Dense(1)(h1)
+co = l.Dense(1)(i)
+
+o = l.Add()([y, co])
+
+
+m = tf.keras.models.Model(inputs = i, outputs = o)
+
+#g.train_generic(m, "c1")
+g.verify_generic(m, "c1")

nn6-cascade.py

@@ -0,0 +1,19 @@
+import tensorflow as tf
+import tensorflow.keras.layers as l
+
+import generic as g
+
+
+i = l.Input(shape = (2,))
+h1 = l.Dense(1000, activation='relu')(i)
+
+y = l.Dense(1)(h1)
+co = l.Dense(1)(i)
+
+o = l.Add()([y, co])
+
+
+m = tf.keras.models.Model(inputs = i, outputs = o)
+
+#g.train_generic(m, "c2")
+g.verify_generic(m, "c2")

nn7-cascade.py

@@ -0,0 +1,24 @@
+import tensorflow as tf
+import tensorflow.keras.layers as l
+
+import generic as g
+
+
+i = l.Input(shape = (2,))
+h1 = l.Dense(100, activation='tanh')(i)
+h2 = l.Dense(100, activation='tanh')(h1)
+h3 = l.Dense(100, activation='tanh')(h2)
+h4 = l.Dense(100, activation='tanh')(h3)
+h5 = l.Dense(100, activation='tanh')(h4)
+h6 = l.Dense(100, activation='tanh')(h5)
+
+y = l.Dense(1)(h6)
+co = l.Dense(1)(i)
+
+o = l.Add()([y, co])
+
+
+m = tf.keras.models.Model(inputs = i, outputs = o)
+
+#g.train_generic(m, "c3")
+g.verify_generic(m, "c3")

nn8-elman.py

@@ -0,0 +1,19 @@
+import tensorflow as tf
+import tensorflow.keras.layers as l
+
+import generic as g
+
+m = tf.keras.models.Sequential([
+    l.Input(shape = (2, 1)),
+    l.SimpleRNN(10, activation = "relu"),
+    l.Dense(1)
+])
+#i = l.Input(shape = (2,))
+#h1 = l.SimpleRNN(10, activation = "relu", input_shape = (2, 1))(i)
+#o = l.Dense(1)(h1)
+
+
+#m = tf.keras.models.Model(inputs = i, outputs = o)
+
+#g.train_generic(m, "r1")
+g.verify_generic(m, "r1")

nn9-elman.py

@@ -0,0 +1,19 @@
+import tensorflow as tf
+import tensorflow.keras.layers as l
+
+import generic as g
+
+m = tf.keras.models.Sequential([
+    l.Input(shape = (2, 1)),
+    l.SimpleRNN(1000, activation = "relu"),
+    l.Dense(1)
+])
+#i = l.Input(shape = (2,))
+#h1 = l.SimpleRNN(10, activation = "relu", input_shape = (2, 1))(i)
+#o = l.Dense(1)(h1)
+
+
+#m = tf.keras.models.Model(inputs = i, outputs = o)
+
+#g.train_generic(m, "r2")
+g.verify_generic(m, "r2")