pysiglib.sig_combine

Contents

pysiglib.sig_combine#

sig_combine(sig1, sig2, dimension, degree, n_jobs=1)[source]#

Combines two truncated signatures of the same degree and dimension into one signature. In particular, let \(x_1, x_2\) be two paths such that the first point of \(x_2\) is the last point of \(x_1\). Let \(S(x_1), S(x_2)\) be the truncated signatures of \(x_1, x_2\) respectively. Then calling this function on \(S(x_1), S(x_2)\) returns the truncated signature of the concatenated path,

\[S(x_1 * x_2) = S(x_1) \otimes S(x_2),\]

where \(x_1 * x_2\) is the concatenation of the two paths \(x_1, x_2\).

Parameters:

  • sig1 (numpy.ndarray | torch.tensor) – The first truncated signature

  • sig2 (numpy.ndarray | torch.tensor) – The second truncated signature. Must have the same degree and dimension as the first.

  • dimension (int) – Dimension of the underlying space, \(d\).

  • degree (int) – Truncation level of the signatures, \(N\)

  • n_jobs (int) – Number of threads to run in parallel. If n_jobs = 1, the computation is run serially. If set to -1, all available threads are used. For n_jobs below -1, (max_threads + 1 + n_jobs) threads are used. For example if n_jobs = -2, all threads but one are used.

Returns:

Combined signature, \(S(x_1 * x_2)\)

Return type:

numpy.ndarray | torch.tensor

Note

Parallelising the computation by setting n_jobs != 1 can be beneficial when the workload is large. However, if the workload is too small, it may be faster to set this to 1 and run the computation serially, due to parallelisation overhead.

Note

Ideally, any array passed to pysiglib.sig_combine should be both contiguous and own its data. If this is not the case, pysiglib.sig_combine will internally create a contiguous copy, which may be inefficient.

Example usage:

import pysiglib

batch_size = 32
length = 100
dimension = 5
degree = 3

X1 = np.random.uniform(size=(batch_size, length, dimension))
X2 = np.random.uniform(size=(batch_size, length, dimension))
X_concat = np.concatenate((X1, X2), axis=1)

X2 = np.concatenate((X1[:, [-1], :], X2), axis=1) # Make sure first pt of X2 is last pt of X1
sig1 = pysiglib.signature(X1, degree)
sig2 = pysiglib.signature(X2, degree)

# The tensor product...
sig_mult = pysiglib.sig_combine(sig1, sig2, dimension, degree)

# ... is the same as the signature of the concatenated path:
sig = pysiglib.signature(X_concat, degree)
Contents