polish,docs; move pairwise() to sklearn_seco.util

evaluation.py

@@ -606,7 +606,7 @@ def plot_speedup(results_log_file: str,
         import seaborn
         seaborn.set(**seaborn_style)
     import matplotlib.pyplot as plt
-    fig: plt.Figure = plt.figure()
+    fig: plt.Figure = plt.figure(figsize=(9, 4))
     ax: plt.Axes = fig.gca(xscale='log', yscale='log')
 
     for other in _OTHER_ALGO:
@@ -625,6 +625,7 @@ def plot_speedup(results_log_file: str,
                t4[('runtime_cv', 'sklearn_seco.Ripper')].values,
                t4['speedup'].values)
 
+    fig.tight_layout()
     if outfile_pattern is not None:
         fig.savefig(outfile_pattern)
     return fig

seco_runtime_scaling.py

@@ -78,6 +78,11 @@ def plot_timings(timings, title=None, figure=None,
     axes = figure.gca(xlabel='n_features', ylabel='time[s]')
     if title is not None:
         axes.set_title(title)
+
+    timings = np.asarray(timings)
+    # sort by n_features so plot lines make sense
+    timings = timings[np.argsort(timings[:, 1])]
+
     n_samples = timings.T[0]
     n_features = timings.T[1]
     tm_min = timings.T[2]

sklearn_seco/abstract.py

@@ -507,6 +507,7 @@ def wrapper_ordering_classes_by_size(estimator):
                 self.base_estimator_ = OneVsOneClassifier(self.base_estimator_,
                                                           n_jobs=self.n_jobs)
             elif self.multi_class_ == "direct":
+                # TODO: if self.multi_class=='direct' (not `None` auto-detect), only assertion prevents binary-only learner to silently learn on multiclass training data
                 self.base_estimator_ = wrapper_ordering_classes_by_size(
                     self.base_estimator_)
             else:

sklearn_seco/common.py

@@ -371,9 +371,10 @@ class RuleContext:
     Methods & Properties provided are:
       - `PN`
       - `X` and `y`
-      - `match_rule`
       - `pn`
+      - `match_rule`
       - `evaluate_rule`
+      - `sort_key`
 
     Fields
     -----

sklearn_seco/concrete.py

@@ -24,16 +24,7 @@
     AbstractSecoImplementation, RuleContext, TheoryContext
 from sklearn_seco.ripper_mdl import \
     data_description_length, relative_description_length
-from sklearn_seco.util import log2
-
-
-def pairwise(iterable):
-    """s -> (s0,s1), (s1,s2), (s2, s3), ..."""
-    # copied from itertools docs
-    from itertools import tee
-    a, b = tee(iterable)
-    next(b, None)
-    return zip(a, b)
+from sklearn_seco.util import log2, pairwise
 
 
 def grow_prune_split(y,

sklearn_seco/tests/datasets.py

@@ -20,7 +20,6 @@
 from typing import Union
 
 import numpy as np
-import pytest
 from sklearn.datasets import make_blobs, make_classification, make_moons
 from sklearn.externals import _arff as arff
 from sklearn.utils import check_random_state, Bunch

sklearn_seco/util.py

@@ -16,6 +16,15 @@ def log2(x: float) -> float:
     return math.log2(x) if x > 0 else 0
 
 
+def pairwise(iterable):
+    """s -> (s0,s1), (s1,s2), (s2, s3), ..."""
+    # copied from itertools docs
+    from itertools import tee
+    a, b = tee(iterable)
+    next(b, None)
+    return zip(a, b)
+
+
 def build_categorical_mask(which_features, n_features: int
                            ) -> np.ndarray or None:
     """:return: A mask array of length `n_features` based on `which_features`.