def view(self, img_only=False):
import ergo.views as views
views.model(self, img_only)
views.roc(self, img_only)
views.stats(self, img_only)
views.history(self, img_only)
views.show(img_only)
def action_explore(argc, argv):
global prj, nrows, ncols, attributes, n_jobs
args = parse_args(argv)
if args.all:
args.pca = True
args.correlations = True
args.stats = True
args.cluster = True
args.D3 = True
if args.workers == -1:
import multiprocessing
n_jobs = multiprocessing.cpu_count()
elif args.workers != 0:
n_jobs = args.workers
log.info("using %d workers" % n_jobs)
if args.nclusters and not args.cluster:
log.warning(
"number of clusters specified but clustering won't be perfomed")
if not (args.pca or args.correlations or args.stats or args.cluster):
log.error("No exploration action was specified")
print("")
parse_args(["-h"])
quit()
prj = Project(args.path)
err = prj.load()
if err is not None:
log.error("error while loading project: %s", err)
quit()
prj.prepare(args.dataset, 0.0, 0.0)
if not prj.dataset.is_flat:
log.error("data exploration can only be applied to flat inputs")
quit()
X, y = prj.dataset.subsample(args.ratio)
nrows, ncols = X.shape
attributes = get_attributes(args.attributes, ncols)
if args.correlations:
log.info("computing correlations of each feature with target")
corr = compute_correlations_with_target(X, y)
print_target_correlation_table(corr)
log.info("computing features crosscorrelation")
corr = calculate_corr(X)
print_correlation_table(corr, min_corr=0.7)
views.correlation_matrix(prj, corr, args.img_only)
if args.pca:
log.info("computing pca")
pca = calculate_pca(X)
log.info("computing pca projection")
views.pca_projection(prj, pca, X, y, False)
if args.D3:
views.pca_projection(prj, pca, X, y, args.D3)
views.pca_explained_variance(prj, pca, args.img_only)
if args.stats:
log.info("computing features stats")
print_stats_table(X)
inertia = False
if args.cluster:
if args.cluster_alg == 'kmeans':
cluster_alg = kmeans_clustering
if not args.nclusters:
args.nclusters = len(set(np.argmax(y, axis=1)))
args.nclusters = int(args.nclusters)
if args.nmaxclusters:
log.info(
"performing inertia analysis with clusters in the range (%d, %d)"
% (args.nclusters, args.nmaxclusters))
inertia = True
n_clusters_analysis(X, args.nmaxclusters, args.nclusters)
else:
log.info("computing kmeans clustering with k=%d" %
args.nclusters)
elif args.cluster_alg == 'dbscan':
cluster_alg = dbscan_clustering
if not args.nclusters:
args.nclusters = 2
log.info("computing dbscan clustering with eps=%f" %
args.nclusters)
if args.nmaxclusters:
log.warning(
"nmax specified but not used. Inertia analysis only available for Kmeans."
)
if not args.pca and not inertia:
log.info("computing pca to plot clusters")
pca = calculate_pca(X)
if not inertia:
ca = cluster_alg(X, args.nclusters)
if len(set(ca.labels_)) == 1:
log.error("clustering failed. Check input parameter.")
quit()
views.plot_clusters(prj, pca, X, y, ca, False)
if args.D3:
views.plot_clusters(prj, pca, X, y, ca, args.D3)
views.show(args.img_only)