def testBuildDummyQNetIncorrectTFVars(self):
tf_vars = [
tf.Variable([[.1], [.2], [.3]], name='weights', dtype=tf.float32)
]
ph_dict = utils.create_placeholders_for_q_net(tf_vars)
with self.assertRaises(AssertionError):
utils.build_dummy_q_net([[3.]], [[2., 1.]], ph_dict, tf_vars)
def testBuildDummyQNetTwoLayers(self):
hidden_weights = np.array([[-.1, .1], [-.2, .2], [-.3, .3]])
hidden_bias = np.array([.3, .3])
output_weights = np.array([[.5], [.5]])
output_bias = np.array([-.05])
tf_vars = [
tf.Variable(hidden_weights,
name='hidden_weights',
dtype=tf.float32),
tf.Variable(hidden_bias, name='hidden_bias', dtype=tf.float32),
tf.Variable(output_weights,
name='output_weights',
dtype=tf.float32),
tf.Variable(output_bias, name='output_bias', dtype=tf.float32)
]
ph_dict = utils.create_placeholders_for_q_net(tf_vars)
output = utils.build_dummy_q_net([[3.]], [[2., 1.]], ph_dict, tf_vars)
with self.test_session() as sess:
self.assertAllClose(
[[.6]],
sess.run(output,
feed_dict={
ph_dict['{}_ph'.format(tf_vars[0].name)]:
hidden_weights,
ph_dict['{}_ph'.format(tf_vars[1].name)]:
hidden_bias,
ph_dict['{}_ph'.format(tf_vars[2].name)]:
output_weights,
ph_dict['{}_ph'.format(tf_vars[3].name)]:
output_bias,
}))
def testBuildDummyQNetSingleLinearUnit(self):
weights = np.array([[.1], [.2], [.3]])
bias = np.array([-.5])
tf_vars = [tf.Variable(weights, name='weights', dtype=tf.float32),
tf.Variable(bias, name='bias', dtype=tf.float32)]
ph_dict = utils.create_placeholders_for_q_net(tf_vars)
output = utils.build_dummy_q_net([[3.]], [[2., 1.]], ph_dict, tf_vars)
with self.test_session() as sess:
self.assertAllClose([[.5]], sess.run(output, feed_dict={
ph_dict['{}_ph'.format(tf_vars[0].name)]: weights,
ph_dict['{}_ph'.format(tf_vars[1].name)]: bias,
}))
def testCreatePlaceholdersForQNet(self):
tf_vars = [tf.Variable([True], name='var1', dtype=tf.bool),
tf.Variable([2, 3], name='var2', dtype=tf.int32),
tf.Variable([4, 5, 6], name='var3', dtype=tf.float32)]
ph_dict = utils.create_placeholders_for_q_net(tf_vars)
self.assertEqual(['var1:0_ph', 'var2:0_ph', 'var3:0_ph'],
list(ph_dict.keys()))
v = ph_dict['var1:0_ph']
self.assertEqual([1], v.shape.as_list())
self.assertEqual(tf.bool, v.dtype)
v = ph_dict['var2:0_ph']
self.assertEqual([2], v.shape.as_list())
self.assertEqual(tf.int32, v.dtype)
v = ph_dict['var3:0_ph']
self.assertEqual([3], v.shape.as_list())
self.assertEqual(tf.float32, v.dtype)