pkg relative imports

* update docs
* explaining constants
* freeze packages for Py2
* fix ransac_segm

Author: Borda

MANIFEST.in

@@ -11,7 +11,6 @@ include LICENSE
 
 # Include the Requirements
 include requirements.txt
-exclude requirements-*.txt
 
 # Include package
 recursive-include imsegm *.py *.pyx

experiments_ovary_centres/run_center_evaluation.py

@@ -156,7 +156,7 @@ def load_center_evaluate(idx_row, df_annot, path_annot, path_visu=None,
         return dict_row
 
     assert all(c in df_annot.columns for c in tl_visu.COLUMNS_POSITION_EGG_ANNOT), \
-        'some required columns %s are missing for %s' % \
+        'some required columns %r are missing for %s' % \
         (tl_visu.COLUMNS_POSITION_EGG_ANNOT, df_annot.columns)
     mask_eggs = estimate_eggs_from_info(df_annot.loc[idx], img.shape[:2])
 

experiments_ovary_detect/run_ellipse_annot_match.py

@@ -83,7 +83,7 @@ def select_optimal_ellipse(idx_row, path_dir_csv, overlap_thr=OVERLAP_THRESHOLD)
     path_csv = os.path.join(path_dir_csv, row['image_name'] + '.csv')
     df_ellipses = pd.read_csv(path_csv, index_col=0)
 
-    pos = row[tl_visu.COLUMNS_POSITION_EGG_ANNOT]
+    pos = row[list[tl_visu.COLUMNS_POSITION_EGG_ANNOT]]
     max_size = 2 * max(pos) + min(pos)
 
     pos_ant = [[row['ant_x'], row['ant_y']]]

imsegm/__init__.py

@@ -5,8 +5,8 @@
 """
 
 try:
-    import imsegm.utilities
-    imsegm.utilities
+    from . import utilities
+    utilities
 except ImportError:
     import traceback
     traceback.print_exc()

imsegm/annotation.py

@@ -15,11 +15,12 @@
 from scipy import interpolate
 
 # sys.path += [os.path.abspath('.'), os.path.abspath('..')]  # Add path to root
-from imsegm.utilities.data_io import io_imread
+from .utilities.data_io import io_imread
 
+#: names of annotated columns
 COLUMNS_POSITION = ('ant_x', 'ant_y', 'post_x', 'post_y', 'lat_x', 'lat_y')
 SLICE_NAME_GROUPING = 'stack_path'
-# set distance in Z axis whether near sliuce may still belong to the same egg
+#: set distance in Z axis whether near slice may still belong to the same egg
 ANNOT_SLICE_DIST_TOL = {  # stage : distance in z-axis
     1: 1,
     2: 2,
@@ -28,11 +29,11 @@
     5: 3,
     6: 0,
 }
-# default colors for particular label
+#: default colors for particular label
 DICT_COLOURS = {
-    0: (0, 0, 255),
-    1: (255, 0, 0),
-    2: (0, 255, 0),
+    0: (0, 0, 255),  # blue
+    1: (255, 0, 0),  # red
+    2: (0, 255, 0),  # green
     3: (255, 229, 0),  # yellow
     4: (142, 68, 173),  # purple
     5: (127, 140, 141),  # gray

imsegm/classification.py

@@ -31,27 +31,37 @@
 except Exception:
     from sklearn.model_selection import GridSearchCV, RandomizedSearchCV
 
-from imsegm.labeling import relabel_max_overlap_unique
-from imsegm.utilities.experiments import WrapExecuteSequence, nb_workers
+from .labeling import relabel_max_overlap_unique
+from .utilities.experiments import WrapExecuteSequence, nb_workers
 
 # NAME_FILE_RESULTS = 'results.csv'
+#: name template forexporting trained classifier (adding classifier name and version)
 TEMPLATE_NAME_CLF = 'classifier_{}.pkl'
+#: default (recommended) classifier for supervised segmentation
 DEFAULT_CLASSIF_NAME = 'RandForest'
+#: default (recommended) clustering for unsupervised segmentation
 DEFAULT_CLUSTERING = 'kMeans'
 # DEFAULT_MIN_NB_SPL = 25
 # nb_workers_CLASSIF_SEARCH = 5
 # NB_CLASSIF_SEARCH_ITER = 250
+#: file name of exported evaluation on feature quality
 NAME_CSV_FEATURES_SELECT = 'feature_selection.csv'
+#: exporting partial results about trained classifier
 NAME_CSV_CLASSIF_CV_SCORES = 'classif_{}_cross-val_scores-{}.csv'
+#: exporting partial results about trained classifier - Receiver Operating Characteristics
 NAME_CSV_CLASSIF_CV_ROC = 'classif_{}_cross-val_ROC-{}.csv'
+#: exporting partial results about trained classifier - Area Under Curve
 NAME_TXT_CLASSIF_CV_AUC = 'classif_{}_cross-val_AUC-{}.txt'
+#: default types of computed metrics
 METRIC_AVERAGES = ('macro', 'weighted')
+#: default computed metrics
 METRIC_SCORING = ('f1_macro', 'accuracy', 'precision_macro', 'recall_macro')
-# rounding unique features, in case to detail precision
+#: rounding unique features, in case to detail precision
 ROUND_UNIQUE_FTS_DIGITS = 3
+#: default number of workers
 NB_WORKERS_SERACH = nb_workers(0.5)
 
-
+#: mapping of metrics names to used functions
 DICT_SCORING = {
     'f1': metrics.f1_score,
     'accuracy': metrics.accuracy_score,

imsegm/descriptors.py

@@ -29,16 +29,24 @@
     print('descriptors: using pure python libraries')
     USE_CYTHON = False
 
+#: define all available statistic computed on superpixels
 NAMES_FEATURE_FLAGS = ('mean', 'std', 'energy', 'median', 'meanGrad')
+#: define sigmas for Lewen-Malik filter bank
 DEFAULT_FILTERS_SIGMAS = (np.sqrt(2), 2, 2 * np.sqrt(2), 4)
+#: define small list/range of sigmas for Lewen-Malik filter bank
 SHORT_FILTERS_SIGMAS = (np.sqrt(2), 2, 4)
+#: define the richest version of computed superpixel features
 FEATURES_SET_ALL = {'color': ('mean', 'std', 'energy', 'median', 'meanGrad'),
                     'tLM': ('mean', 'std', 'energy', 'median', 'meanGrad')}
+#: define basic color features for supepixels
 FEATURES_SET_COLOR = {'color': ('mean', 'std', 'energy')}
+#: define basic texture features (complete LM filter bank) for supepixels
 FEATURES_SET_TEXTURE = {'tLM': ('mean', 'std', 'energy')}
+#: define basic color features for (small LM filter bank) supepixels
 FEATURES_SET_TEXTURE_SHORT = {'tLM_short': ('mean', 'std', 'energy')}
+#: define circular diamters for computing label histogram
 HIST_CIRCLE_DIAGONALS = (10, 20, 30, 40, 50)
-# maxila reposnse is bounded by fix number to preven overflowing
+#: maximal response is bounded by fix number to prevent overflowing (for LM filer bank)
 MAX_SIGNAL_RESPONSE = 1.e6
 
 # Wavelets:

imsegm/ellipse_fitting.py

@@ -10,17 +10,21 @@
 
 from skimage.measure import fit as sk_fit
 # from skimage.measure.fit import EllipseModel  # fix in future skimage>0.13.0
-from imsegm.utilities.drawing import ellipse
-from imsegm.descriptors import (reduce_close_points, compute_ray_features_segm_2d,
-                                reconstruct_ray_features_2d)
-from imsegm.superpixels import (segment_slic_img2d, superpixel_centers,
-                                make_graph_segm_connect_grid2d_conn4)
-
-INIT_MASK_BORDER = 50.
+from .utilities.drawing import ellipse
+from .descriptors import (
+    reduce_close_points, compute_ray_features_segm_2d, reconstruct_ray_features_2d)
+from .superpixels import (
+    segment_slic_img2d, superpixel_centers, make_graph_segm_connect_grid2d_conn4)
+
+# INIT_MASK_BORDER = 50.
+#: define minimal size of estimated ellipse
 MIN_ELLIPSE_DAIM = 25.
+#: define maximal Figure size in larger dimension
 MAX_FIGURE_SIZE = 14
-SEGM_OVERLAP = 0.5
+# SEGM_OVERLAP = 0.5  # define transparency for overlapping two images
+#: smoothing background with morphological operation
 STRUC_ELEM_BG = 15
+#: smoothing foreground with morphological operation
 STRUC_ELEM_FG = 5
 
 
@@ -207,21 +211,21 @@ def ransac_segm(points, model_class, points_all, weights, labels, table_prob,
     best_inliers = None
 
     if isinstance(min_samples, float):
-        if not (0 <= min_samples <= 1):
-            raise ValueError("`min_samples` as ration must be in range (0, 1)")
+        if not (0 < min_samples <= 1):
+            raise ValueError("`min_samples` as ration must be in range (0, 1]")
         min_samples = int(min_samples * len(points))
-    if min_samples < 0:
-        raise ValueError("`min_samples` must be greater than zero")
+    if not (0 < min_samples <= len(points)):
+        raise ValueError("`min_samples` must be in range (0, <nb-samples>]")
 
     if max_trials < 0:
         raise ValueError("`max_trials` must be greater than zero")
 
-    # make sure points is list and not tuple, so it can be modified below
+    # make sure points is ndarray and not tuple/list, so it can be modified below
     points = np.array(points)
 
     for _ in range(max_trials):
         # choose random sample set
-        random_idxs = np.random.randint(0, len(points), min_samples)
+        random_idxs = np.random.choice(len(points), min_samples, replace=False)
         samples = points[random_idxs]
         # for d in points:
         #     samples.append(d[random_idxs])

imsegm/graph_cuts.py

@@ -12,18 +12,21 @@
 from sklearn import metrics, preprocessing
 from sklearn import pipeline, cluster, mixture, decomposition
 
-from imsegm.utilities.drawing import (
+from .utilities.drawing import (
     draw_graphcut_unary_cost_segments, draw_graphcut_weighted_edges, draw_color_labeling)
-from imsegm.superpixels import (
+from .superpixels import (
     make_graph_segm_connect_grid2d_conn4, make_graph_segm_connect_grid3d_conn6, superpixel_centers)
-from imsegm.descriptors import compute_selected_features_img2d
+from .descriptors import compute_selected_features_img2d
 
+#: define munber of iteration in Grap-Cut optimization
 DEFAULT_GC_ITERATIONS = 25
-COEF_INT_CONVERSION = 1e6
-DEBUG_NB_SHOW_SAMPLES = 15
+# COEF_INT_CONVERSION = 1e6
+# DEBUG_NB_SHOW_SAMPLES = 15
+#: define minimal value of unary (being a class) term in Graph-Cut
 MIN_UNARY_PROB = 0.01
+#: define maximal value of parwise (smoothness) term in Graph-Cut
 MAX_PAIRWISE_COST = 1e5
-# max is this value and min is inverse (1 / val)
+#: max is this value and min is inverse (1 / val)
 MIN_MAX_EDGE_WEIGHT = 1e3
 
 

imsegm/labeling.py

@@ -10,7 +10,7 @@
 from scipy import ndimage
 import skimage.segmentation as sk_segm
 
-from imsegm.utilities.data_io import get_image2d_boundary_color
+from .utilities.data_io import get_image2d_boundary_color
 
 
 def neighbour_connect4(seg, label, pos):