dimensionReduction.py

Go to the documentation of this file.

def doIt(X, method, ncomponents, nneighbors, njobs, rstate, params):
 
    import importlib.util
 
    # if numpy, scipy and sklearn are not found
    not_found = False
 
    # check if modules are installed
    for mod in ["numpy", "scipy", "sklearn"]:
        if importlib.util.find_spec(mod) is None:
            not_found = True
            print("[DimensionReduction] Python error: " + mod + " module not found.")
 
    if not_found:
        # at least one module is not installed, aborting...
        return 0
 
    import sklearn
    from sklearn import manifold
    from sklearn import decomposition
    import numpy as np
    from sys import platform
    from packaging import version
 
    #if platform == "darwin":
    #    sklearn.utils.parallel_backend("threading")
 
    if rstate > 0:
        np.random.seed(0)
 
    try:
        if method == 0:
            seParams = params[0]
            if seParams[2] == "None":
                se = manifold.SpectralEmbedding(
                    n_components=ncomponents,
                    affinity=seParams[0],
                    gamma=seParams[1],
                    eigen_solver=None,
                    n_neighbors=nneighbors,
                    n_jobs=njobs,
                )
            else:
                se = manifold.SpectralEmbedding(
                    n_components=ncomponents,
                    affinity=seParams[0],
                    gamma=seParams[1],
                    eigen_solver=seParams[2],
                    n_neighbors=nneighbors,
                    n_jobs=njobs,
                )
            Y = se.fit_transform(X)
        elif method == 1:
            lleParams = params[1]
            lle = manifold.LocallyLinearEmbedding(
                n_neighbors=nneighbors,
                n_components=ncomponents,
                reg=lleParams[0],
                eigen_solver=lleParams[1],
                tol=lleParams[2],
                max_iter=lleParams[3],
                method=lleParams[4],
                hessian_tol=lleParams[5],
                modified_tol=lleParams[6],
                neighbors_algorithm=lleParams[7],
                n_jobs=njobs,
            )
            Y = lle.fit_transform(X)
        elif method == 2:
            mdsParams = params[2]
            mds = manifold.MDS(
                n_components=ncomponents,
                metric=mdsParams[0],
                n_init=mdsParams[1],
                max_iter=mdsParams[2],
                verbose=mdsParams[3],
                eps=mdsParams[4],
                dissimilarity=mdsParams[5],
                n_jobs=njobs,
            )
            Y = mds.fit_transform(X)
        elif method == 3:
            tsneParams = params[3]
            if version.parse(sklearn.__version__) >= version.parse("1.7.0"):
                tsne = manifold.TSNE(
                    n_components=ncomponents,
                    perplexity=tsneParams[0],
                    early_exaggeration=tsneParams[1],
                    learning_rate=tsneParams[2],
                    max_iter=tsneParams[3],
                    n_iter_without_progress=tsneParams[4],
                    min_grad_norm=tsneParams[5],
                    metric=tsneParams[6],
                    init=tsneParams[7],
                    verbose=tsneParams[8],
                    method=tsneParams[9],
                    angle=tsneParams[10],
                )
            else:
                tsne = manifold.TSNE(
                    n_components=ncomponents,
                    perplexity=tsneParams[0],
                    early_exaggeration=tsneParams[1],
                    learning_rate=tsneParams[2],
                    n_iter=tsneParams[3],
                    n_iter_without_progress=tsneParams[4],
                    min_grad_norm=tsneParams[5],
                    metric=tsneParams[6],
                    init=tsneParams[7],
                    verbose=tsneParams[8],
                    method=tsneParams[9],
                    angle=tsneParams[10],
                )
            Y = tsne.fit_transform(X)
        elif method == 4:
            isoParams = params[4]
            iso = manifold.Isomap(
                n_neighbors=nneighbors,
                n_components=ncomponents,
                eigen_solver=isoParams[0],
                tol=isoParams[1],
                max_iter=isoParams[2],
                path_method=isoParams[3],
                neighbors_algorithm=isoParams[4],
                metric=isoParams[5],
                n_jobs=njobs,
            )
            Y = iso.fit_transform(X)
        elif method == 5:
            pcaParams = params[5]
            if pcaParams[4] == "auto":
                pca = decomposition.PCA(
                    n_components=ncomponents,
                    copy=pcaParams[0],
                    whiten=pcaParams[1],
                    svd_solver=pcaParams[2],
                    tol=pcaParams[3],
                    iterated_power="auto",
                )
            else:
                pca = decomposition.PCA(
                    n_components=ncomponents,
                    copy=pcaParams[0],
                    whiten=pcaParams[1],
                    svd_solver=pcaParams[2],
                    tol=pcaParams[3],
                    iterated_power=int(pcaParams[4]),
                )
            Y = pca.fit_transform(X)
 
        L = [Y.shape[0], Y.shape[1], np.ravel(Y, "F")]
        return L
    except Exception as inst:
        print(
            "[DimensionReduction] Error: unexpected behaviour detected in the python script."
        )
        print(type(inst))  # the exception instance
        print(inst.args)  # arguments stored in .args
        print(inst)
        return []