985a8dba993da323aeb1656169521e5f868372a4,basenji/rnn.py,RNN,test,#RNN#Any#Any#Any#Any#Any#,688

Before Change

        si = 0

        // setup feed dict for dropout
        fd = {self.is_training:False}
        for li in range(self.cnn_layers):
            fd[self.cnn_dropout_ph[li]] = 0
        for li in range(self.dcnn_layers):
            fd[self.dcnn_dropout_ph[li]] = 0
        for li in range(self.rnn_layers):
            fd[self.rnn_dropout_ph[li]] = 0

        // get first batch
        Xb, Yb, NAb, Nb = batcher.next()

        while Xb is not None:
            // update feed dict

After Change

        self.save_reprs = job.get("save_reprs", False)


    def test(self, sess, batcher, rc_avg=False, return_preds=False, down_sample=1):
        """ Compute model accuracy on a test set.

        Args:
          sess:         TensorFlow session
          batcher:      Batcher object to provide data
          rc_avg:       Average predictions from the forward and reverse complement sequences
          return_preds: Bool indicating whether to return predictions
          down_sample:  Int specifying to consider uniformly spaced sampled positions

        Returns:
          mean_loss:    Mean loss across targets
          mean_r2:      Mean R^2 across targets
          preds:        Predictions
        """

        batch_losses = []

        // determine non-buffer region
        buf_start = self.batch_buffer // self.target_pool
        buf_end = (self.batch_length - self.batch_buffer) // self.target_pool
        buf_len = buf_end - buf_start

        // uniformly sample indexes
        ds_indexes = np.arange(0, buf_len, down_sample)

        // initialize prediction and target arrays
        preds = np.zeros((batcher.num_seqs, len(ds_indexes), self.num_targets), dtype="float16")
        targets = np.zeros((batcher.num_seqs, len(ds_indexes), self.num_targets), dtype="float16")
        targets_na = np.zeros((batcher.num_seqs, len(ds_indexes)), dtype="bool")
        si = 0

        // setup feed dict
        fd = self.set_mode("test")

        // get first batch
        Xb, Yb, NAb, Nb = batcher.next()

Instances