Python 3.x support for hierarchical topic models (#833)

* * Python 3.x support for hierarchical topic models
* Fix bug in hARTM.load(...) -- models saved with hARTM.save(...) didn't load before

* Re-structure imports in hierarchy_utils.py

python/artm/hierarchy_utils.py

@@ -4,13 +4,14 @@
 import uuid
 import copy
 import numpy as np
+import pandas
 import os.path
 import os
 import pickle
 import glob
 import warnings
-from pandas import DataFrame
-import pandas
+
+from six.moves import range
 
 
 class hARTM(object):
@@ -287,7 +288,7 @@ def del_level(self, level_idx):
         if level_idx == -1:
             del self._levels[-1]
             return
-        for _ in xrange(level_idx, len(self._levels)):
+        for _ in range(level_idx, len(self._levels)):
             del self._levels[-1]
 
     def get_level(self, level_idx):
@@ -363,7 +364,7 @@ def save(self, path):
             level.save(os.path.join(path, "level" +
                                     str(level_idx) + "_pwt.model"), model_name="p_wt")
         info = {"parent_level_weight": [
-            level.phi_batch_weight for level in self._levels[1:]]}
+            level.parent_level_weight for level in self._levels[1:]]}
         with open(os.path.join(path, "info.dump"), "wb") as fout:
             pickle.dump(info, fout)
 
@@ -386,14 +387,14 @@ def load(self, path):
         info_filename = glob.glob(os.path.join(path, "info.dump"))
         if len(info_filename) != 1:
             raise ValueError("Given path is not hARTM safe")
-        with open(info_filename[0]) as fin:
+        with open(info_filename[0], "rb") as fin:
             info = pickle.load(fin)
         model_filenames = glob.glob(os.path.join(path, "*.model"))
         if len({len(info["parent_level_weight"]) + 1, len(model_filenames) / 2}) > 1:
             raise ValueError("Given path is not hARTM safe")
         self._levels = []
         sorted_model_filenames = sorted(model_filenames)
-        for level_idx in xrange(len(model_filenames) / 2):
+        for level_idx in range(len(model_filenames) // 2):
             if not len(self._levels):
                 model = artm.ARTM(num_topics=1,
                                   seed=self._get_seed(level_idx),
@@ -407,10 +408,10 @@ def load(self, path):
                                    num_topics=1,
                                    seed=self._get_seed(level_idx),
                                    **self._common_models_args)
-            filename = sorted_model_filenames[2 * level_idx]
-            model.load(filename, "n_wt")
             filename = sorted_model_filenames[2 * level_idx + 1]
             model.load(filename, "p_wt")
+            filename = sorted_model_filenames[2 * level_idx]
+            model.load(filename, "n_wt")
             config = model.master._config
             config.opt_for_avx = False
             model.master._lib.ArtmReconfigureMasterModel(
@@ -650,9 +651,9 @@ def get_theta(self, topic_names=None):
         _, nd_array = self.master.get_theta_matrix(topic_names=use_topic_names)
 
         titles_list = [item_title for item_title in theta_info.item_title]
-        theta_data_frame = DataFrame(data=nd_array.transpose(),
-                                     columns=titles_list,
-                                     index=use_topic_names)
+        theta_data_frame = pandas.DataFrame(data=nd_array.transpose(),
+                                            columns=titles_list,
+                                            index=use_topic_names)
         item_idxs = np.logical_not(
             theta_data_frame.columns.isin(self._parent_model.topic_names))
         theta_data_frame = theta_data_frame.drop(