API Reference
pyhuge.huge
huge(
x,
lambda_=None,
nlambda=None,
lambda_min_ratio=None,
method="mb",
scr=None,
scr_num=None,
cov_output=False,
sym="or",
verbose=True,
backend="native",
*,
input_type="auto",
) -> HugeResult
Native entry for graph-path estimation.
input_type="auto" preserves symmetry-based detection. Use "data" for a
square symmetric observation matrix, or "covariance" to require a square
covariance/correlation matrix.
For glasso with no explicit lambda, auto-detected covariance input matches R's
historical diagonal-sensitive default scale; explicit covariance routing uses
the corrected off-diagonal scale. Glasso may accept an indefinite
pairwise estimate only when the native solver certifies a positive-definite,
internally consistent result. CT and TIGER continue to require positive
semidefiniteness.
pyhuge.test
test(require_runtime=False) -> dict
Environment probe for native runtime.
Returned keys include:
python_importnumpyscipysklearn(compatibility field; not required for runtime)native_extensionruntimerpy2(compatibility field)
Wrapper shortcuts
huge_mb(x, lambda_=None, nlambda=None, lambda_min_ratio=None,
scr=None, scr_num=None, sym="or", verbose=True, backend="native",
*, input_type="auto")
huge_glasso(x, lambda_=None, nlambda=None, lambda_min_ratio=None,
scr=None, cov_output=False, verbose=True, backend="native",
*, input_type="auto")
huge_ct(x, lambda_=None, nlambda=None, lambda_min_ratio=None,
verbose=True, backend="native", *, input_type="auto")
huge_tiger(x, lambda_=None, nlambda=None, lambda_min_ratio=None,
sym="or", verbose=True, backend="native", *,
input_type="auto")
These call huge(...) with the method fixed; arguments match the
corresponding subset of huge().
pyhuge.huge_select
huge_select(
est,
criterion=None,
ebic_gamma=0.5,
stars_thresh=0.1,
stars_subsample_ratio=None,
rep_num=20,
n_jobs=1,
verbose=True,
backend="native",
) -> HugeSelectResult
Model selection on HugeResult. With criterion="stars", n_jobs > 1 fits
the subsamplings in a thread pool (the native solvers release the GIL);
results are identical to the serial path. Mirrors num.cores in the R
package. Each fit may also start OpenMP or BLAS threads; use n_jobs=1 when
a bounded thread budget matters. TIGER currently supports RIC selection, not StARS: subsample TIGER
fits can certify different path prefixes, and no common-prefix protocol is
exposed yet. When criterion=None, defaults match R: RIC for MB/TIGER,
StARS for CT, and EBIC for graphical lasso. Parameters are validated only
when their criterion uses them. StARS requires a non-increasing
est.lambda_path; tied values are allowed.
pyhuge.huge_npn
huge_npn(x, npn_func="shrinkage", verbose=True) -> numpy.ndarray
Native nonparanormal transformation.
pyhuge.huge_generator
huge_generator(
n=200,
d=50,
graph="random",
v=None,
u=None,
g=None,
prob=None,
vis=False,
verbose=True,
random_state=None,
) -> HugeGeneratorResult
Native synthetic data generator.
pyhuge.huge_inference
huge_inference(
data,
t,
adj,
alpha=0.05,
type_="Gaussian",
method="score",
) -> HugeInferenceResult
Native edge-wise inference approximation. Data must have at least two rows and no constant columns; Nonparanormal inference also requires at least two variables. The precision-like matrix must have a finite, positive diagonal.
pyhuge.huge_roc
huge_roc(path, theta, verbose=True, plot=False) -> HugeRocResult
Native ROC metrics over graph path. The truth matrix must contain at least one edge and one absent off-diagonal edge; otherwise ROC/AUC is undefined.
pyhuge.huge_stockdata
huge_stockdata() -> HugeStockDataResult
Loads packaged stock dataset (1258 x 452 matrix + 452 x 3 info table).
Summaries
huge_summary(fit: HugeResult) -> HugeSummary
huge_select_summary(sel: HugeSelectResult) -> HugeSelectSummary
Plot helpers
huge_plot_sparsity(fit, ax=None, show_points=True)
huge_plot_roc(roc, ax=None)
huge_plot_graph_matrix(fit, index=-1, ax=None)
huge_plot_network(fit, index=-1, ax=None, layout="spring",
with_labels=False, node_size=120.0,
node_color="#c44e52", edge_color="#4d4d4d",
min_abs_weight=0.0)
huge_plot(g, epsflag=False, graph_name="default", cur_num=1, location=None)
Dataclasses
HugeResultHugeSelectResultHugeGeneratorResultHugeInferenceResultHugeRocResultHugeStockDataResultHugeSummaryHugeSelectSummary
Exception
PyHugeError: raised for validation failures or missing native dependencies.