29eac269527a4466dfef282374aed49ce66d9bfb,chainer_/models/nasnet.py,NASNet,__init__,#NASNet#Any#Any#Any#Any#Any#,991

Before Change

                        in_channels = out_channels
                self.features.add(stage)

            self.features.add(nn.Activation("relu"))
            self.features.add(nn.AvgPool2D(
                pool_size=7,
                strides=1))

After Change

                return_two=False,
                first_ordinals=1,
                last_ordinals=2)
            with self.features.init_scope():
                setattr(self.features, "init_block", NASNetInitBlock(
                    in_channels=in_channels,
                    out_channels=init_block_channels))
                in_channels = init_block_channels

                out_channels = stem_blocks_channels[0]
                setattr(self.features, "stem1_unit", Stem1Unit(
                    in_channels=in_channels,
                    out_channels=out_channels))
                prev_in_channels = in_channels
                in_channels = out_channels

                out_channels = stem_blocks_channels[1]
                setattr(self.features, "stem2_unit", Stem2Unit(
                    in_channels=in_channels,
                    prev_in_channels=prev_in_channels,
                    out_channels=out_channels))
                prev_in_channels = in_channels
                in_channels = out_channels

                for i, channels_per_stage in enumerate(channels):
                    stage = nasnet_dual_path_sequential()
                    with stage.init_scope():
                        for j, out_channels in enumerate(channels_per_stage):
                            if (j == 0) and (i != 0):
                                unit = ReductionUnit
                            elif ((i == 0) and (j == 0)) or ((i != 0) and (j == 1)):
                                unit = FirstUnit
                            else:
                                unit = NormalUnit
                            setattr(stage, "unit{}".format(j + 1), unit(
                                in_channels=in_channels,
                                prev_in_channels=prev_in_channels,
                                out_channels=out_channels))
                            prev_in_channels = in_channels
                            in_channels = out_channels
                    setattr(self.features, "stage{}".format(i + 1), stage)

                setattr(self.features, "final_activ", F.relu)
                setattr(self.features, "final_pool", partial(
                    F.average_pooling_2d,
                    ksize=7,
                    stride=1))

            self.output = SimpleSequential()
            with self.output.init_scope():
                setattr(self.output, "flatten", partial(
                    F.reshape,

Instances