Sampling

Helpers to down- or upsample time series data.

ecgan.preprocessing.sampling.transpose(input, dim0, dim1) → Tensor

Returns a tensor that is a transposed version of input. The given dimensions dim0 and dim1 are swapped.

The resulting out tensor shares its underlying storage with the input tensor, so changing the content of one would change the content of the other.

Parameters

  • input (Tensor) -- the input tensor.

  • dim0 (int) -- the first dimension to be transposed

  • dim1 (int) -- the second dimension to be transposed

Example:

>>> x = torch.randn(2, 3)
>>> x
tensor([[ 1.0028, -0.9893,  0.5809],
        [-0.1669,  0.7299,  0.4942]])
>>> torch.transpose(x, 0, 1)
tensor([[ 1.0028, -0.1669],
        [-0.9893,  0.7299],
        [ 0.5809,  0.4942]])
ecgan.preprocessing.sampling.downsampling_fixed_sample_rate(data, sample_rate)[source]

Downsample dataset by returning every sample_rate-th value in every dimension.

Parameters

  • data (ndarray) -- Tensor to be downsampled. Shape: (seq_len, channels).

  • sample_rate (int) -- The fixed sample rate used to retain every sample_rate-th value.

Return type

ndarray

Returns

The downsampled series.

ecgan.preprocessing.sampling.downsample_largest_triangle_three_buckets(data, threshold)[source]

Downsample the data according to LTTB.

Parameters

  • data (ndarray) -- The unsampled data.

  • threshold (int) -- The LTTB threshold (target size).

Return type

ndarray

Returns

The downsampled data.

ecgan.preprocessing.sampling.interpolate(data, target_frequency, interpolation_strategy=None)[source]

Force an incoming multivariate series to conform to a fixed frequency using PyTorch.

Required if measuring devices use a different sampling frequency. More information on sampling rates: See here. Requires 3D input for most interpolation strategies.

Parameters

  • data (ndarray) -- Input series which shall be upsampled. Shape: (seq_len, num_channels).

  • target_frequency (int) -- Desired output frequency.

  • interpolation_strategy (Optional[str]) -- Strategy according to https://pytorch.org/docs/stable/nn.functional.html

Return type

ndarray

Returns

(NumPy) Tensor with the upsampled values.

ecgan.preprocessing.sampling.resample(data, target_rate, algorithm=SamplingAlgorithm.LTTB, interpolation_strategy=None)[source]

Sample data according to the specified SamplingAlgorithm.

Parameters

  • data (ndarray) -- The data that shall be sampled.

  • algorithm (SamplingAlgorithm) -- The sampling algorithm.

  • target_rate (int) -- Has to be set for a fixed sampling rate.

  • interpolation_strategy (Optional[str]) -- According to the PyTorch interpolation strategies. Only required for SamplingAlgorithm.INTERPOLATE.

Return type

ndarray

Returns

The resampled data.