Fixes to spade modules regarding the integration of an updated version of fim (#140)

* Fixes to spade modules which integrate an updated version of `fim` and updated corresponding tests

Author: pietroquaglio

elephant/spade.py

@@ -7,6 +7,15 @@
 SPADE analysis can be applied as demonstrated in this short toy example of 10
 artificial spike trains of exhibiting fully synchronous events of order 10.
 
+This modules relies on the implementation of the fp-growth algorithm contained
+in the file fim.so which can be found here (http://www.borgelt.net/pyfim.html)
+and should be available in the spade_src folder (elephant/spade_src/).
+If the fim.so module is not present in the correct location or cannot be 
+imported (only available for linux OS) SPADE will make use of a python 
+implementation of the fast fca algorithm contained in 
+elephant/spade_src/fast_fca.py, which is about 10 times slower.
+
+
 import elephant.spade
 import elephant.spike_train_generation
 import quantities as pq
@@ -56,6 +65,7 @@
 import time
 import quantities as pq
 import warnings
+warnings.simplefilter('once', UserWarning)
 try:
     from mpi4py import MPI  # for parallelized routines
     HAVE_MPI = True
@@ -65,9 +75,11 @@
 try:
     from elephant.spade_src import fim
     HAVE_FIM = True
-    # raise ImportError
 except ImportError:  # pragma: no cover
     HAVE_FIM = False
+    warnings.warn(
+        'fim.so not found in elephant/spade_src folder,' +
+        'you are using the python implementation of fast fca')
 from elephant.spade_src import fast_fca
 
 
@@ -444,7 +456,8 @@ def concepts_mining(data, binsize, winlen, min_spikes=2, min_occ=2,
     # By default, set the maximum pattern size to the maximum number of
     # spikes in a window
     if max_spikes is None:
-        max_spikes = int(np.max(np.sum(rel_matrix, axis=1)))
+        max_spikes = np.max((int(np.max(np.sum(rel_matrix, axis=1))),
+                            min_spikes + 1))
     # By default, set maximum number of occurrences to number of non-empty
     # windows
     if max_occ is None:
@@ -537,7 +550,7 @@ def _build_context(binary_matrix, winlen):
         rel_matrix[w, :] = times
         # appending to the transactions spike idx (fast_fca input) of the
         # current window (fpgrowth input)
-        transactions.append(attributes[times])
+        transactions.append(list(attributes[times]))
     # Return context and rel_matrix
     return context, transactions, rel_matrix
 
@@ -624,26 +637,30 @@ def _fpgrowth(transactions, min_c=2, min_z=2, max_z=None,
     '''
     # By default, set the maximum pattern size to the number of spiketrains
     if max_z is None:
-        max_z = np.max([len(tr) for tr in transactions]) + 1
+        max_z = np.max((np.max([len(tr) for tr in transactions]), min_z + 1))
     # By default set maximum number of data to number of bins
     if max_c is None:
-        max_c = len(transactions) + 1
-    if report != '#' or min_neu > 1:
+        max_c = len(transactions)
+    if min_neu >= 1:
         if min_neu < 1:
             raise AttributeError('min_neu must be an integer >=1')
         # Inizializing outputs
         concepts = []
         spec_matrix = np.zeros((max_z, max_c))
         spectrum = []
         # Mining the data with fpgrowth algorithm
-        fpgrowth_output = fim.fpgrowth(
-            tracts=transactions,
-            target=target,
-            supp=-min_c,
-            min=min_z,
-            max=max_z,
-            report='a',
-            algo='s')
+        if np.unique(transactions, return_counts=True)[1][0] == len(
+                transactions):
+            fpgrowth_output = [(tuple(transactions[0]), len(transactions))]
+        else:
+            fpgrowth_output = fim.fpgrowth(
+                tracts=transactions,
+                target=target,
+                supp=-min_c,
+                zmin=min_z,
+                zmax=max_z,
+                report='a',
+                algo='s')
         # Applying min/max conditions and computing extent (window positions)
         fpgrowth_output = list(filter(
             lambda c: _fpgrowth_filter(
@@ -665,16 +682,6 @@ def _fpgrowth(transactions, min_c=2, min_z=2, max_z=None,
             return spectrum
         else:
             return concepts
-    elif report == '#' and min_neu == 1:
-        spectrum = fim.fpgrowth(
-            tracts=transactions,
-            target=target,
-            supp=-min_c,
-            min=min_z,
-            max=max_z,
-            report=report,
-            algo='s')
-        return spectrum
     else:
         raise AttributeError('min_neu must be an integer >=1')
 
@@ -687,7 +694,7 @@ def _fpgrowth_filter(concept, winlen, max_c, min_neu):
     keep_concepts = len(
         np.unique(
             np.array(
-                concept[0]) // winlen)) >= min_neu and concept[1][0] <= max_c
+                concept[0]) // winlen)) >= min_neu and concept[1] <= max_c
     return keep_concepts
 
 

elephant/test/test_spade.py

@@ -6,23 +6,22 @@
 """
 from __future__ import division
 import unittest
-import os
-import warnings
 
 import neo
 import numpy as np
-from numpy.testing.utils import assert_array_almost_equal, assert_array_equal
+from numpy.testing.utils import assert_array_equal
 import quantities as pq
-import elephant.spike_train_generation as stg
 import elephant.spade as spade
 import elephant.conversion as conv
 import elephant.spike_train_generation as stg
+
 try:
-    import fim
+    from elephant.spade_src import fim
     HAVE_FIM = True
 except ImportError:
     HAVE_FIM = False
 
+
 class SpadeTestCase(unittest.TestCase):
     def setUp(self):
         # Spade parameters