Fix evaluate().explore() in Python 3

Fixes #1866

* Use `list()` to get the same behavior from filter on 2 & 3
* Use `six.iteritems()` to get `iteritems()` in 2 and `items()` in 3
* Removes roc_curve and confusion_matrix from the serialized data (it
  was inadvertently added in #1891, and causes serialization errors)
* Add unit test coverage to prevent this type of issue

src/unity/python/turicreate/data_structures/image.py

@@ -233,6 +233,11 @@ def show(self):
 
         """
         from ..visualization._plot import _target
+
+        # Suppress visualization output if 'none' target is set
+        if _target == 'none':
+            return
+
         try:
             img = self._to_pil_image()
             try:

src/unity/python/turicreate/data_structures/sframe.py

@@ -4479,6 +4479,12 @@ def explore(self, title=None):
         if sys.platform != 'darwin' and sys.platform != 'linux2' and sys.platform != 'linux':
             raise NotImplementedError('Visualization is currently supported only on macOS and Linux.')
 
+
+        # Suppress visualization output if 'none' target is set
+        from ..visualization._plot import _target
+        if _target == 'none':
+            return
+
         path_to_client = _get_client_app_path()
 
         if title is None:

src/unity/python/turicreate/test/test_image_classifier.py

@@ -242,6 +242,12 @@ def test_save_and_load(self):
             self.test_list_fields()
             print("List fields passed")
 
+    def test_evaluate_explore(self):
+        # Run the explore method and make sure we don't throw an exception.
+        # This will test the JSON serialization logic.
+        tc.visualization.set_target('none')
+        evaluation = self.model.evaluate(data)
+        evaluation.explore()
 
 class ImageClassifierSqueezeNetTest(ImageClassifierTest):
     @classmethod

src/unity/python/turicreate/toolkits/image_classifier/_evaluation.py

@@ -12,6 +12,7 @@
 from ...visualization import _get_client_app_path, _focus_client_app
 
 import subprocess as __subprocess
+import six as _six
 from six.moves import _thread
 
 import json as _json
@@ -28,10 +29,10 @@ def __init__(self, obj = {}):
   def _get_eval_json(self):
     evaluation_dictionary = dict()
 
-    for key, value in self.data.iteritems():
+    for key, value in _six.iteritems(self.data):
       if (isinstance(value, float) or isinstance(value, int)) and _math.isnan(value):
         continue
-      if (key is "test_data"):
+      if (key is "test_data" or key is "confusion_matrix" or key is "roc_curve"):
         continue
       evaluation_dictionary[key] = value
 
@@ -48,7 +49,12 @@ def _get_eval_json(self):
     return str(_json.dumps({ "evaluation_spec": evaluation_dictionary }, allow_nan = False))
 
   def explore(self):
-    _thread.start_new_thread(_start_process, (self._get_eval_json()+"\n", self.data["test_data"], self, ))
+    params = (self._get_eval_json()+"\n", self.data["test_data"], self, )
+    # Suppress visualization output if 'none' target is set
+    from ...visualization._plot import _target
+    if _target == 'none':
+        return
+    _thread.start_new_thread(_start_process, params)
 
 
 def _get_data_spec(filters, start, length, row_type, mat_type, sframe, evaluation):

src/unity/python/turicreate/toolkits/image_classifier/image_classifier.py

@@ -741,8 +741,8 @@ def hclusterSort(vectors, dist_fn):
 
                 excluding_names = [min_dist['from']['name'], min_dist['to']['name']]
 
-                vecs = filter(lambda v: v['name'] not in excluding_names, vecs)
-                distances = filter(lambda dist: (dist['from']['name'] not in excluding_names) and (dist['to']['name'] not in excluding_names), distances)
+                vecs = list(filter(lambda v: v['name'] not in excluding_names, vecs))
+                distances = list(filter(lambda dist: (dist['from']['name'] not in excluding_names) and (dist['to']['name'] not in excluding_names), distances))
 
                 for v in vecs:
                     total = 0

src/unity/python/turicreate/visualization/_plot.py

@@ -46,8 +46,8 @@ def _run_cmdline(command):
 
 def set_target(target='auto'):
     """
-    Sets the target for visualizations launched with the `show` method. If
-    unset, or if target is not provided, defaults to 'auto'.
+    Sets the target for visualizations launched with the `show` or `explore`
+    methods. If unset, or if target is not provided, defaults to 'auto'.
 
     Notes
     -----
@@ -63,10 +63,11 @@ def set_target(target='auto'):
         * 'auto': display plot output inline when in Jupyter Notebook, and
           otherwise launch a native GUI window.
         * 'gui': always launch a native GUI window.
+        * 'none': prevent all visualizations from being displayed.
     """
     global _target
-    if target not in ['auto', 'gui']:
-        raise ValueError("Expected target to be one of: 'auto', 'gui'.")
+    if target not in ['auto', 'gui', 'none']:
+        raise ValueError("Expected target to be one of: 'auto', 'gui', 'none'.")
     _target = target
 
 
@@ -121,6 +122,11 @@ def show(self):
 
         """
         global _target
+
+        # Suppress visualization output if 'none' target is set
+        if _target == 'none':
+            return
+
         display = False
         try:
             if _target == 'auto' and \