ca4c92fe1b53086373cfca478560df06b03701d8,examples/atari_dqn/convnet.py,ConvNet,train_on_batch,#ConvNet#Any#Any#,60

Before Change

        actions = x[1].astype(np.int)

        t = self.q.predict(x[0])
        for i in xrange(t.shape[0]):
            t[i, actions[i]] = y[i]

        loss = self.q.train_on_batch(x[0], t, **fit_params)
        summary = tf.Summary(value=[tf.Summary.Value(tag="loss",
                                                     simple_value=loss), ])
        self.writer.add_summary(summary)

After Change

        return self._session.run(self.q, feed_dict={self._x: x})

    def train_on_batch(self, x, y, **fit_params):
        self._session.run(self._train_step,
                          feed_dict={self._x: x[0],
                                     self._action: x[1].ravel().astype(
                                         np.uint8),
                                     self._target_q: y})

    def set_weights(self, weights):
        with tf.variable_scope(self._name):
            w = tf.get_collection(ops.GraphKeys.TRAINABLE_VARIABLES,

Instances