Transformer (Pairwise Panel)

FlatDist

Panel distance or kernel from applying tabular trafo to flattened time series.

Applies the wrapped tabular distance or kernel to flattened series. Flattening is done to a 2D numpy array of shape (n_instances, (n_vars, n_timepts))

Formal details (for real valued objects, mixed typed rows in analogy): Let \(d:\mathbb{R}^k \times \mathbb{R}^{k}\rightarrow \mathbb{R}\) be the pairwise function in transformer, when applied to k-vectors (here, \(d\) could be a distance function or a kernel function). Let \(x_1, \dots, x_N\in \mathbb{R}^{n \times \ell}\), \(y_1, \dots y_M \in \mathbb{R}^{n \times \ell}\) be collections of matrices, representing time series panel valued inputs X and X2, as follows: \(x_i\) is the i-th instance in X, and \(x_{i, j\ell}\) is the j-th time point, \ell-th variable of X. Analogous for \(y\) and X2. Let \(f:\mathbb{R}^{n \times \ell} \rightarrow \mathbb{R}^{n \cdot \ell}\) be the mapping that flattens matrices by column-first lexicographical ordering, and assume \(k = n \cdot \ell\).

Then, transform(X, X2) returns the \((N \times M)\) matrix with \((i, j)\)-th entry \(d\left(f(x_i), f(y_j)\right)\).

Schnellstart

python

from sktime.dists_kernels.compose_tab_to_panel import FlatDist

estimator = FlatDist(transformer)

Parameter(1)

transformer: pairwise transformer of BasePairwiseTransformer scitype, or: callable np.ndarray (n_samples, d) x (n_samples, d) -> (n_samples x n_samples)

Beispiele

Euclidean distance between time series of equal length, considered as vectors
>>> from sktime.dists_kernels import FlatDist, ScipyDist
>>> euc_tsdist = FlatDist (ScipyDist ()) Gaussian kernel between time series of equal length, considered as vectors
>>> from sklearn.gaussian_process.kernels import RBF
>>> flat_gaussian_tskernel = FlatDist (RBF ())