-
Notifications
You must be signed in to change notification settings - Fork 3
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Add tests for diagnostic, make the functions more robust to types, fi…
…x a small bug in constraint
- Loading branch information
1 parent
ca811b2
commit 99803f1
Showing
4 changed files
with
170 additions
and
18 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,112 @@ | ||
| import adelie as ad | ||
| import numpy as np | ||
| import pytest | ||
| import scipy | ||
|
|
||
|
|
||
| def generate_data(n, p, L, K, beta_type, dtype, seed): | ||
| np.random.seed(seed) | ||
| X = np.random.normal(0, 1, (n, p)).astype(dtype) | ||
| y = X @ np.random.normal(0, 1, (p, K)) + np.random.normal(0, 1, (n, K)) | ||
| y = y.astype(dtype) | ||
| betas = np.random.uniform(-1, 1, (L, p*K)).astype(dtype) | ||
| if beta_type == "sparse": | ||
| betas = scipy.sparse.csr_matrix(betas, dtype=dtype) | ||
| if K == 1: | ||
| y = y.squeeze(axis=1) | ||
| intercepts = np.random.normal(0, 1, (L,)).astype(dtype) | ||
| else: | ||
| intercepts = np.random.normal(0, 1, (L, K)).astype(dtype) | ||
| lmdas = np.sort(np.random.uniform(0, 1, L))[::-1] | ||
|
|
||
| return { | ||
| "X": X, | ||
| "y": y, | ||
| "betas": betas, | ||
| "intercepts": intercepts, | ||
| "lmdas": lmdas, | ||
| } | ||
|
|
||
|
|
||
| def predict(X, betas, intercepts): | ||
| p = X.shape[1] | ||
| L = betas.shape[0] | ||
| K = 1 if len(intercepts.shape) == 1 else intercepts.shape[1] | ||
| if K != 1: | ||
| if isinstance(betas, scipy.sparse.csr_matrix): | ||
| betas = betas.toarray() | ||
| betas = betas.reshape((L, p, K)) | ||
| Xbetas = np.einsum("ij,ljk->lik", X, betas) | ||
| else: | ||
| Xbetas = betas @ X.T | ||
| return intercepts[:, None] + Xbetas | ||
|
|
||
|
|
||
| @pytest.mark.filterwarnings("ignore: Detected matrix to be C-contiguous.") | ||
| @pytest.mark.parametrize("dtype", [np.float32, np.float64]) | ||
| @pytest.mark.parametrize("beta_type", ["dense", "sparse"]) | ||
| @pytest.mark.parametrize("K", [1, 3]) | ||
| @pytest.mark.parametrize("L", [1, 2]) | ||
| @pytest.mark.parametrize("p", [1, 10]) | ||
| @pytest.mark.parametrize("n", [2, 20]) | ||
| def test_predict(n, p, L, K, beta_type, dtype, seed=0): | ||
| data = generate_data(n, p, L, K, beta_type, dtype, seed) | ||
| X = data["X"] | ||
| betas = data["betas"] | ||
| intercepts = data["intercepts"] | ||
| actual = ad.diagnostic.predict(X, betas, intercepts) | ||
| expected = predict(X, betas, intercepts) | ||
|
|
||
| assert np.allclose(actual, expected, atol=1e-6) | ||
|
|
||
|
|
||
| @pytest.mark.filterwarnings("ignore: Detected matrix to be C-contiguous.") | ||
| @pytest.mark.parametrize("dtype", [np.float32, np.float64]) | ||
| @pytest.mark.parametrize("beta_type", ["dense", "sparse"]) | ||
| @pytest.mark.parametrize("K", [1, 3]) | ||
| @pytest.mark.parametrize("L", [1, 2]) | ||
| @pytest.mark.parametrize("p", [1, 10]) | ||
| @pytest.mark.parametrize("n", [2, 20]) | ||
| def test_objective(n, p, L, K, beta_type, dtype, seed=0): | ||
| def _objective(X, glm, betas, intercepts, lmdas): | ||
| etas = predict(X, betas, intercepts) | ||
| losses = np.array([glm.loss(eta) - glm.loss_full() for eta in etas]) | ||
| if isinstance(betas, scipy.sparse.csr_matrix): | ||
| betas = betas.toarray() | ||
| if K == 1: | ||
| penalty = np.sum(np.abs(betas), axis=-1) | ||
| else: | ||
| betas = betas.reshape((L, p, K)) | ||
| penalty = np.sum(np.linalg.norm(betas, axis=-1), axis=-1) | ||
| return losses + lmdas * np.sqrt(K) * penalty | ||
|
|
||
| data = generate_data(n, p, L, K, beta_type, dtype, seed) | ||
| X = data["X"] | ||
| y = data["y"] | ||
| betas = data["betas"] | ||
| intercepts = data["intercepts"] | ||
| lmdas = data["lmdas"] | ||
| if K == 1: | ||
| glm = ad.glm.gaussian(y) | ||
| else: | ||
| glm = ad.glm.multigaussian(y) | ||
|
|
||
| actual = ad.diagnostic.objective(X, glm, betas, intercepts, lmdas) | ||
| expected = _objective(X, glm, betas, intercepts, lmdas) | ||
|
|
||
| assert np.allclose(actual, expected, atol=1e-6) | ||
|
|
||
|
|
||
| @pytest.mark.filterwarnings("ignore: Detected matrix to be C-contiguous.") | ||
| @pytest.mark.parametrize("dtype", [np.float32, np.float64]) | ||
| def test_diagnostic(dtype, seed=0): | ||
| n = 10 | ||
| p = 10 | ||
| np.random.seed(seed) | ||
| X = np.random.normal(0, 1, (n, p)).astype(dtype) | ||
| y = X @ np.random.normal(0, 1, p) + np.random.normal(0, 1, n) | ||
| y = y.astype(dtype) | ||
| constraints = [None] * p | ||
| constraints[0] = ad.constraint.lower(np.full(1, -1, dtype=dtype)) | ||
| state = ad.grpnet(X, ad.glm.gaussian(y), progress_bar=False) | ||
| ad.diagnostic.diagnostic(state) |