d58f2f8913722e0c95e665ab442da0c12a1766ae,basenji/rnn.py,RNN,build,#RNN#Any#,15

Before Change

        """

        // throw away a buffer region on the side
        outputs = outputs[self.batch_buffer:-self.batch_buffer]

        // make final predictions
        preds_length = []
        for li in range(self.batch_length - 2*self.batch_buffer):

After Change

        sq_diff = tf.squared_difference(self.preds, self.targets[:,self.batch_buffer:self.batch_length-self.batch_buffer,:])

        // set any NaN"s to zero
        nan_indexes = tf.is_nan(sq_diff)
        tens0 = tf.zeros_like(sq_diff)
        sq_diff = tf.select(nan_indexes, tens0, sq_diff)

        // take the mean
        self.loss_op = tf.reduce_mean(sq_diff)

        // define optimization
        if self.optimization == "adam":
            opt = tf.train.AdamOptimizer(self.learning_rate, self.adam_beta1, self.adam_beta2)

Instances