samuelkim314 / deepsymreg Goto Github PK

import numpy as np
import tensorflow as tf
from utils import functions, pretty_print
from utils.symbolic_network import SymbolicNetL0
from utils.regularization import l12_smooth


funcs = functions.default_func
x_dim = 100
x = np.random.rand(100, 1)
y = x ** 2

# Set up TensorFlow graph for the EQL network
x_placeholder = tf.placeholder(shape=(None, x_dim), dtype=tf.float32)
sym = SymbolicNetL0(symbolic_depth=2, funcs=funcs)
y_hat = sym(x_placeholder)

# Set up loss function with L0.5 loss
mse = tf.losses.mean_squared_error(labels=y, predictions=y_hat)
loss = mse + 1e-2 * l12_smooth(sym.get_weights())

# Set up TensorFlow graph for training
opt = tf.train.RMSPropOptimizer()
train = opt.minimize(loss)

# Random data for a simple function
x = np.random.rand(100, 1)
y = x ** 2

# Training
with tf.Session as sess:
  sess.run(tf.global_variables_initializer())
  for i in range(1000):
    sess.run(train, feed_dict={x_placeholder: x})

  # Print out the expression
  weights = sess.run(sym.get_weights())
  expr = pretty_print.network(weights, funcs, ['x'])
  print(expr)

# Initialize weights for last layer (without activation functions)
self.output_weight = tf.Variable(tf.random_uniform(shape=(self.symbolic_layers[-1].n_funcs, 1)))

# Initialize weights for last layer (without activation functions)
self.output_weight = tf.Variable(tf.random_uniform(shape=(self.symbolic_layers[-1].n_funcs, 4)))