Siegel spaces #179
Siegel spaces #179
Conversation
…d distance, Takagi factorization, Cayley transformation and other utils functions for operating with complex symmetric matrices
|
|
||
| name = "Bounded Domain" | ||
|
|
||
| def __init__(self, metric: str = SiegelMetricType.RIEMANNIAN.value, rank: int = None): |
There was a problem hiding this comment.
With Enum you can do the following
from enum import Enum
class SiegelMetricType(Enum):
"""Supported metric types for Siegel Spaces"""
RIEMANNIAN = "riem"
FINSLER_ONE = "fone"
FINSLER_INFINITY = "finf"
FINSLER_MINIMUM = "fmin"
WEIGHTED_SUM = "wsum"
and later
<SiegelMetricType.RIEMANNIAN: 'riem'>
or
SiegelMetricType(SiegelMetricType.RIEMANNIAN)
<SiegelMetricType.RIEMANNIAN: 'riem'>
so no need in from_str method
|
|
||
| name = "Bounded Domain" | ||
|
|
||
| def __init__(self, metric: str = SiegelMetricType.RIEMANNIAN.value, rank: int = None): |
There was a problem hiding this comment.
| def __init__(self, metric: str = SiegelMetricType.RIEMANNIAN.value, rank: int = None): | |
| def __init__(self, metric: str = SiegelMetricType.RIEMANNIAN, rank: int = None): |
| FINSLER_MINIMUM = "fmin" | ||
| WEIGHTED_SUM = "wsum" | ||
|
|
||
| @staticmethod |
There was a problem hiding this comment.
Removed, thanks!
| @classmethod | ||
| def get(cls, type_str: str, rank: int): | ||
| metrics_map = { | ||
| SiegelMetricType.RIEMANNIAN.value: RiemannianMetric, |
There was a problem hiding this comment.
| SiegelMetricType.RIEMANNIAN.value: RiemannianMetric, | |
| SiegelMetricType.RIEMANNIAN: RiemannianMetric, |
No need in .value
|
|
||
| @classmethod | ||
| def get(cls, type_str: str, rank: int): | ||
| metrics_map = { |
There was a problem hiding this comment.
This can be a static class variable
There was a problem hiding this comment.
Since I cannot have the children classes declared in a static class variable in the parent class, I extracted the creation logic in a SiegelMetricFactory
| return seed | ||
|
|
||
|
|
||
| @pytest.fixture(scope="module", params=[torch.complex128, torch.complex64]) |
There was a problem hiding this comment.
maybe add float dtype to the test suit?
There was a problem hiding this comment.
We use the Cayley Transform to move from one model (Upper half or Bounded domain) to the other and vice versa, thus we only define it for complex matrices.
|
As linters say, |
…Type instead of a string * Moved creation logic for Siegel metrics from class to * Corrected tests according to these changes * Added entries in CHANGELOG and README * Run black geoopt tests for formatting
|
|
Fixed issues with pydocstyle and lint |
|
Waiting for the tests pass now |
|
Great! It's merged! |
Implementation of two models of the Siegel Space.
Points in these manifolds are complex symmetric matrices.
Reference:
Federico López, Beatrice Pozzetti, Steve Trettel, Michael Strube, Anna Wienhard.
"Symmetric Spaces for Graph Embeddings: A Finsler-Riemannian Approach", 2021.