Use sgkit.distarray for PCA

tomwhite · tomwhite · commit 6e6392a1941a · 2024-11-18T11:13:31.000Z
diff --git a/.github/workflows/cubed.yml b/.github/workflows/cubed.yml
@@ -30,4 +30,4 @@ jobs:
 
     - name: Test with pytest
       run: |
-        pytest -v sgkit/tests/test_{aggregation,association,hwe}.py -k 'test_count_call_alleles or test_gwas_linear_regression or test_hwep or test_sample_stats or (test_count_variant_alleles and not test_count_variant_alleles__chunked[call_genotype]) or (test_variant_stats and not test_variant_stats__chunks[chunks2-False])' --use-cubed
+        pytest -v sgkit/tests/test_{aggregation,association,hwe,pca}.py -k 'test_count_call_alleles or test_gwas_linear_regression or test_hwep or test_sample_stats or (test_count_variant_alleles and not test_count_variant_alleles__chunked[call_genotype]) or (test_variant_stats and not test_variant_stats__chunks[chunks2-False]) or (test_pca__array_backend and tsqr)' --use-cubed
diff --git a/sgkit/stats/pca.py b/sgkit/stats/pca.py
@@ -1,20 +1,20 @@
 from typing import Any, Optional, Union
 
-import dask.array as da
 import numpy as np
 import xarray as xr
-from dask_ml.decomposition import TruncatedSVD
 from sklearn.base import BaseEstimator
 from sklearn.pipeline import Pipeline
 from typing_extensions import Literal
 from xarray import DataArray, Dataset
 
+import sgkit.distarray as da
 from sgkit import variables
 
 from ..typing import ArrayLike, DType, RandomStateType
 from ..utils import conditional_merge_datasets
 from .aggregation import count_call_alleles
 from .preprocessing import PattersonScaler
+from .truncated_svd import TruncatedSVD
 
 
 def pca_est(
diff --git a/sgkit/stats/preprocessing.py b/sgkit/stats/preprocessing.py
@@ -1,11 +1,11 @@
 from typing import Hashable, Optional
 
-import dask.array as da
 import numpy as np
 import xarray as xr
 from sklearn.base import BaseEstimator, TransformerMixin
 from xarray import Dataset
 
+import sgkit.distarray as da
 from sgkit import variables
 
 from ..typing import ArrayLike
diff --git a/sgkit/stats/truncated_svd.py b/sgkit/stats/truncated_svd.py
@@ -0,0 +1,248 @@
+from dask.utils import has_keyword
+from sklearn.base import BaseEstimator, TransformerMixin
+
+import sgkit.distarray as da
+
+# Based on the implementation in Dask-ML, with minor changes to support the
+# array API so it can work with both Dask and Cubed.
+
+
+class TruncatedSVD(BaseEstimator, TransformerMixin):
+    def __init__(
+        self,
+        n_components=2,
+        algorithm="tsqr",
+        n_iter=5,
+        random_state=None,
+        tol=0.0,
+        compute=True,
+    ):
+        """Dimensionality reduction using truncated SVD (aka LSA).
+
+        This transformer performs linear dimensionality reduction by means of
+        truncated singular value decomposition (SVD). Contrary to PCA, this
+        estimator does not center the data before computing the singular value
+        decomposition.
+
+        Parameters
+        ----------
+        n_components : int, default = 2
+            Desired dimensionality of output data.
+            Must be less than or equal to the number of features.
+            The default value is useful for visualization.
+
+        algorithm : {'tsqr', 'randomized'}
+            SVD solver to use. Both use the `tsqr` (for "tall-and-skinny QR")
+            algorithm internally. 'randomized' uses an approximate algorithm
+            that is faster, but not exact. See the References for more.
+
+        n_iter : int, optional (default 0)
+            Number of power iterations, useful when the singular values
+            decay slowly. Error decreases exponentially as n_power_iter
+            increases. In practice, set n_power_iter <= 4.
+
+        random_state : int, RandomState instance or None, optional
+            If int, random_state is the seed used by the random number
+            generator;
+            If RandomState instance, random_state is the random number
+            generator;
+            If None, the random number generator is the RandomState instance
+            used by `np.random`.
+
+        tol : float, optional
+            Ignored.
+
+        compute : bool
+            Whether or not SVD results should be computed
+            eagerly, by default True.
+
+        Attributes
+        ----------
+        components_ : array, shape (n_components, n_features)
+
+        explained_variance_ : array, shape (n_components,)
+            The variance of the training samples transformed by a projection to
+            each component.
+
+        explained_variance_ratio_ : array, shape (n_components,)
+            Percentage of variance explained by each of the selected
+            components.
+
+        singular_values_ : array, shape (n_components,)
+            The singular values corresponding to each of the selected
+            components. The singular values are equal to the 2-norms of the
+            ``n_components`` variables in the lower-dimensional space.
+
+        See Also
+        --------
+        dask.array.linalg.tsqr
+        dask.array.linalg.svd_compressed
+
+        References
+        ----------
+
+        Direct QR factorizations for tall-and-skinny matrices in
+        MapReduce architectures.
+        A. Benson, D. Gleich, and J. Demmel.
+        IEEE International Conference on Big Data, 2013.
+        http://arxiv.org/abs/1301.1071
+
+        Notes
+        -----
+        SVD suffers from a problem called "sign indeterminacy", which means
+        the sign of the ``components_`` and the output from transform depend on
+        the algorithm and random state. To work around this, fit instances of
+        this class to data once, then keep the instance around to do
+        transformations.
+
+        .. warning::
+
+           The implementation currently does not support sparse matrices.
+
+        Examples
+        --------
+        >>> from dask_ml.decomposition import TruncatedSVD
+        >>> import dask.array as da
+        >>> X = da.random.normal(size=(1000, 20), chunks=(100, 20))
+        >>> svd = TruncatedSVD(n_components=5, n_iter=3, random_state=42)
+        >>> svd.fit(X)  # doctest: +NORMALIZE_WHITESPACE
+        TruncatedSVD(algorithm='tsqr', n_components=5, n_iter=3,
+                     random_state=42, tol=0.0)
+
+        >>> print(svd.explained_variance_ratio_)  # doctest: +ELLIPSIS
+        [0.06386323 0.06176776 0.05901293 0.0576399  0.05726607]
+        >>> print(svd.explained_variance_ratio_.sum())  # doctest: +ELLIPSIS
+        0.299...
+        >>> print(svd.singular_values_)  # doctest: +ELLIPSIS
+        array([35.92469517, 35.32922121, 34.53368856, 34.138..., 34.013...])
+
+        Note that ``transform`` returns a ``dask.Array``.
+
+        >>> svd.transform(X)
+        dask.array<sum-agg, shape=(1000, 5), dtype=float64, chunksize=(100, 5)>
+        """
+        self.algorithm = algorithm
+        self.n_components = n_components
+        self.n_iter = n_iter
+        self.random_state = random_state
+        self.tol = tol
+        self.compute = compute
+
+    def fit(self, X, y=None):
+        """Fit truncated SVD on training data X
+
+        Parameters
+        ----------
+        X : array-like, shape (n_samples, n_features)
+            Training data.
+
+        y : Ignored
+
+        Returns
+        -------
+        self : object
+            Returns the transformer object.
+        """
+        self.fit_transform(X)
+        return self
+
+    def _check_array(self, X):
+        if self.n_components >= X.shape[1]:
+            raise ValueError(
+                "n_components must be < n_features; "
+                "got {} >= {}".format(self.n_components, X.shape[1])
+            )
+        return X
+
+    def fit_transform(self, X, y=None):
+        """Fit model to X and perform dimensionality reduction on X.
+
+        Parameters
+        ----------
+        X : array-like, shape (n_samples, n_features)
+            Training data.
+
+        y : Ignored
+
+        Returns
+        -------
+        X_new : array, shape (n_samples, n_components)
+            Reduced version of X. This will always be a dense array, of the
+            same type as the input array. If ``X`` was a ``dask.array``, then
+            ``X_new`` will be a ``dask.array`` with the same chunks along the
+            first dimension.
+        """
+        X = self._check_array(X)
+        if self.algorithm not in {"tsqr", "randomized"}:
+            raise ValueError(
+                "`algorithm` must be 'tsqr' or 'randomized', not '{}'".format(
+                    self.algorithm
+                )
+            )
+        if self.algorithm == "tsqr":
+            if has_keyword(da.linalg.svd, "full_matrices"):
+                u, s, v = da.linalg.svd(X, full_matrices=False)
+            else:
+                u, s, v = da.linalg.svd(X)
+            u = u[:, : self.n_components]
+            s = s[: self.n_components]
+            v = v[: self.n_components]
+        else:
+            u, s, v = da.linalg.svd_compressed(
+                X, self.n_components, n_power_iter=self.n_iter, seed=self.random_state
+            )
+
+        X_transformed = u * s
+        explained_var = da.var(X_transformed, axis=0)
+        full_var = da.var(X, axis=0)
+        full_var = da.sum(full_var)
+        explained_variance_ratio = explained_var / full_var
+
+        if self.compute:
+            v, explained_var, explained_variance_ratio, s = da.compute(
+                v, explained_var, explained_variance_ratio, s
+            )
+        self.components_ = v
+        self.explained_variance_ = explained_var
+        self.explained_variance_ratio_ = explained_variance_ratio
+        self.singular_values_ = s
+        self.n_features_in_ = X.shape[1]
+        return X_transformed
+
+    def transform(self, X, y=None):
+        """Perform dimensionality reduction on X.
+
+        Parameters
+        ----------
+        X : array-like, shape (n_samples, n_features)
+             Data to be transformed.
+
+        y : Ignored
+
+        Returns
+        -------
+        X_new : array, shape (n_samples, n_components)
+            Reduced version of X. This will always be a dense array, of the
+            same type as the input array. If ``X`` was a ``dask.array``, then
+            ``X_new`` will be a ``dask.array`` with the same chunks along the
+            first dimension.
+        """
+        return X @ self.components_.T
+
+    def inverse_transform(self, X):
+        """Transform X back to its original space.
+
+        Returns an array X_original whose transform would be X.
+
+        Parameters
+        ----------
+        X : array-like, shape (n_samples, n_components)
+            New data.
+
+        Returns
+        -------
+        X_original : array, shape (n_samples, n_features)
+            Note that this is always a dense array.
+        """
+        # X = check_array(X)
+        return X @ self.components_
diff --git a/sgkit/tests/test_pca.py b/sgkit/tests/test_pca.py
@@ -1,12 +1,12 @@
 from typing import Any, Optional
 
 import allel
-import dask.array as da
 import numpy as np
 import pytest
 import xarray as xr
 from xarray import Dataset
 
+import sgkit.distarray as da
 from sgkit.stats import pca
 from sgkit.stats.pca import count_call_alternate_alleles
 from sgkit.testing import simulate_genotype_call_dataset
