## PyTorch ## Down/up sample 主讲人：龙良曲 ## Outline - Pooling - upsample ReLU ## Downsample |A|B| |---|---| |C|D| Scaling factor : 2 |A|A|B|B| |---|---|---|---| |A|A|B|B| |C|C|D|D| |C|C|D|D| ## Max Size([1, 16, 7, 7]) In [36]: out=F.avg_pool2d(x,2,stride=2) Out[37]: torch.Size([1, 16, 7, 7]) ## upsample |A|B| |---|---| |C|D| Scaling factor : 2 |A|A|B|B| |---|---|---|---| |A|A|B|B| |C|C|D|D| |C|C|D|D|

00:00:00 0 2000-01-01 00:03:00 6 2000-01-01 00:06:00 15 2000-01-01 00:09:00 15 Freq: 3T, dtype: int64 Upsample the series into 30 second bins. >>> series.resample('30S')[0:5] #select first 5 rows 2000-01-01 NaN 2000-01-01 00:01:00 1 2000-01-01 00:01:30 NaN 2000-01-01 00:02:00 2 Freq: 30S, dtype: float64 Upsample the series into 30 second bins and fill the NaN values using the pad method. >>> series.resample('30S' 00:00:30 0 2000-01-01 00:01:00 1 2000-01-01 00:01:30 1 2000-01-01 00:02:00 2 Freq: 30S, dtype: int64 Upsample the series into 30 second bins and fill the NaN values using the bfill method. >>> series.resample('30S'

2011-02-28 3 2011-03-31 4 Freq: M, dtype: int64 Note: In the new API, you can either downsample OR upsample. The prior implementation would allow you to pass an aggregator function (like mean) even though converted to higher frequency periods. 19.10.2 Up Sampling For upsampling, you can specify a way to upsample and the limit parameter to interpolate over the gaps that are created: # from secondly to every 2000-01-01 00:00:00 0.0 2000-01-01 00:01:00 NaN 2000-01-01 00:02:00 2.0 2000-01-01 00:03:00 3.0 Upsample the series into 30 second bins. >>> df.asfreq(freq='30S') s 2000-01-01 00:00:00 0.0 2000-01-01 00:00:30

2011-02-28 3 2011-03-31 4 Freq: M, dtype: int64 Note: In the new API, you can either downsample OR upsample. The prior implementation would allow you to pass an aggregator function (like mean) even though converted to higher frequency periods. 19.10.2 Upsampling For upsampling, you can specify a way to upsample and the limit parameter to interpolate over the gaps that are created: # from secondly to every 2000-01-01 00:00:00 0.0 2000-01-01 00:01:00 NaN 2000-01-01 00:02:00 2.0 2000-01-01 00:03:00 3.0 Upsample the series into 30 second bins. >>> df.asfreq(freq='30S') s 2000-01-01 00:00:00 0.0 2000-01-01 00:00:30

2011-02-28 3 2011-03-31 4 Freq: M, dtype: int64 Note: In the new API, you can either downsample OR upsample. The prior implementation would allow you to pass an aggregator function (like mean) even though converted to higher frequency periods. ## Up Sampling For upsampling, you can specify a way to upsample and the limit parameter to interpolate over the gaps that are created: # from secondly to every 00:00:00 0 2000-01-01 00:03:00 6 2000-01-01 00:06:00 15 2000-01-01 00:09:00 15 Freq: 3T, dtype: int64 Upsample the series into 30 second bins. >>> series.resample('30S').asfreq() [0:5] # select

periods are converted to higher frequency periods. Upsampling For upsampling, you can specify a way to upsample and the limit parameter to interpolate over the gaps that are created: # from secondly to every 2000-01-01 00:00:00 0.0 2000-01-01 00:01:00 NaN 2000-01-01 00:02:00 2.0 2000-01-01 00:03:00 3.0 Upsample the series into 30 second bins. >>> df.asfreq(freq='30S') s 2000-01-01 00:00:00 0.0 2000-01-01 00:00:30 2000-01-01 00:01:30 NaN 2000-01-01 00:02:00 2.0 2000-01-01 00:02:30 NaN 2000-01-01 00:03:00 3.0 Upsample again, providing a fill value. >>> df.asfreq(freq='30S', fill_value=9.0) s 2000-01-01 00:00:00 0

) ) def get_conv_upsampling_block(filters): """ It creates a block to upsample i.e. increase the spatial dimensions of the inputs. Upsampling is required to ensure that the model

periods are converted to higher frequency periods. Upsampling For upsampling, you can specify a way to upsample and the limit parameter to interpolate over the gaps that are created: # from secondly to every 2000-01-01 00:00:00 0.0 2000-01-01 00:01:00 NaN 2000-01-01 00:02:00 2.0 2000-01-01 00:03:00 3.0 Upsample the series into 30 second bins. >>> df.asfreq(freq='30S') s 2000-01-01 00:00:00 0.0 2000-01-01 00:00:30 2000-01-01 00:01:30 NaN 2000-01-01 00:02:00 2.0 2000-01-01 00:02:30 NaN 2000-01-01 00:03:00 3.0 Upsample again, providing a fill value. >>> df.asfreq(freq='30S', fill_value=9.0) s 2000-01-01 00:00:00 0

periods are converted to higher frequency periods. Upsampling For upsampling, you can specify a way to upsample and the limit parameter to interpolate over the gaps that are created: # from secondly to every 2000-01-01 00:00:00 0.0 2000-01-01 00:01:00 NaN 2000-01-01 00:02:00 2.0 2000-01-01 00:03:00 3.0 Upsample the series into 30 second bins. >>> df.asfreq(freq='30S') s 2000-01-01 00:00:00 0.0 2000-01-01 00:00:30 2000-01-01 00:01:30 NaN 2000-01-01 00:02:00 2.0 2000-01-01 00:02:30 NaN 2000-01-01 00:03:00 3.0 Upsample again, providing a fill value. >>> df.asfreq(freq='30S', fill_value=9.0) s 2000-01-01 00:00:00 0

periods are converted to higher frequency periods. Upsampling For upsampling, you can specify a way to upsample and the limit parameter to interpolate over the gaps that are created: # from secondly to every 2000-01-01 00:00:00 0.0 2000-01-01 00:01:00 NaN 2000-01-01 00:02:00 2.0 2000-01-01 00:03:00 3.0 Upsample the series into 30 second bins. >>> df.asfreq(freq='30S') s 2000-01-01 00:00:00 0.0 2000-01-01 00:00:30 2000-01-01 00:01:30 NaN 2000-01-01 00:02:00 2.0 2000-01-01 00:02:30 NaN 2000-01-01 00:03:00 3.0 Upsample again, providing a fill value. >>> df.asfreq(freq='30S', fill_value=9.0) s 2000-01-01 00:00:00 0