Merge pull request #101 from tyarkoni/various-improvements

Major improvements

.gitignore

@@ -1,6 +1,7 @@
 *.pyc
 *.DS_Store
 *~
+benchmarks/
 build/
 dist
 dist/*

pliers/config.py

@@ -0,0 +1,5 @@
+
+cache_converters = False
+cache_filters = False
+cache_extractors = False
+log_transformations = True

pliers/converters/__init__.py

@@ -1,37 +1,25 @@
 from abc import ABCMeta, abstractmethod, abstractproperty
 from pliers.transformers import Transformer, CollectionStimMixin
 from six import with_metaclass
-from tempfile import mkdtemp
-from joblib import Memory
+from pliers.utils import memory
 import importlib
+from types import GeneratorType
+from pliers import config
 
 
-cachedir = mkdtemp()
-memory = Memory(cachedir=cachedir, verbose=0)
-
-__all__ = ['api', 'audio', 'google', 'image', 'video', 'multistep']
+__all__ = ['api', 'audio', 'google', 'image', 'iterators', 'video', 'multistep']
 
 
 class Converter(with_metaclass(ABCMeta, Transformer)):
     ''' Base class for Converters.'''
 
     def __init__(self):
         super(Converter, self).__init__()
-        self.convert = memory.cache(self.convert)
+        if config.cache_converters:
+            self.transform = memory.cache(self.transform)
 
     def convert(self, stim, *args, **kwargs):
-        new_stim = self._convert(stim, *args, **kwargs)
-        if new_stim.name is None:
-            new_stim.name = stim.name
-        else:
-            new_stim.name = stim.name + '_' + new_stim.name
-        if isinstance(new_stim, CollectionStimMixin):
-            for s in new_stim:
-                if s.name is None:
-                    s.name = stim.name
-                else:
-                    s.name = stim.name + '_' + s.name
-        return new_stim
+        return self.transform(stim, *args, **kwargs)
 
     @abstractmethod
     def _convert(self, stim):
@@ -42,7 +30,18 @@ def _output_type(self):
         pass
 
     def _transform(self, stim, *args, **kwargs):
-        return self.convert(stim, *args, **kwargs)
+        new_stim = self._convert(stim, *args, **kwargs)
+        if isinstance(new_stim, (list, tuple, GeneratorType)):
+            return new_stim
+        if new_stim.name is None:
+            new_stim.name = stim.name
+        else:
+            new_stim.name = stim.name + '->' + new_stim.name
+        if isinstance(new_stim, CollectionStimMixin):
+            for s in new_stim:
+                if s.name is None:
+                    s.name = stim.name
+        return new_stim
 
 
 def get_converter(in_type, out_type):

pliers/converters/api.py

@@ -48,24 +48,14 @@ def _convert(self, audio):
 
 class WitTranscriptionConverter(SpeechRecognitionAPIConverter):
 
-    @property
-    def environ_key(self):
-        return 'WIT_AI_API_KEY'
-
-    @property
-    def recognize_method(self):
-        return 'recognize_wit'
+    environ_key = 'WIT_AI_API_KEY'
+    recognize_method ='recognize_wit'
 
 
 class GoogleSpeechAPIConverter(SpeechRecognitionAPIConverter):
 
-    @property
-    def environ_key(self):
-        return 'GOOGLE_API_KEY'
-
-    @property
-    def recognize_method(self):
-        return 'recognize_google'
+    environ_key = 'GOOGLE_APPLICATION_CREDENTIALS'
+    recognize_method = 'recognize_google_cloud'
 
 
 class IBMSpeechAPIConverter(AudioToTextConverter):

pliers/converters/google.py

@@ -8,7 +8,6 @@ class GoogleVisionAPITextConverter(GoogleVisionAPITransformer, ImageToTextConver
     request_type = 'TEXT_DETECTION'
     response_object = 'textAnnotations'
 
-
     def __init__(self, handle_annotations='first', **kwargs):
         super(GoogleVisionAPITextConverter, self).__init__(**kwargs)
         self.handle_annotations = handle_annotations

pliers/converters/iterators.py

@@ -0,0 +1,34 @@
+from pliers.stimuli.video import VideoStim, DerivedVideoStim
+from pliers.stimuli.image import ImageStim
+from pliers.stimuli.text import ComplexTextStim, TextStim
+from pliers.converters import Converter
+
+
+class StimCollectionIterator(Converter):
+
+    def _convert(self, stim):
+        for element in stim:
+            yield element
+
+
+class VideoFrameIterator(StimCollectionIterator):
+
+    _input_type = VideoStim
+    _output_type = ImageStim
+
+
+class DerivedVideoFrameIterator(StimCollectionIterator):
+
+    # TODO: use VideoFrameIterator for both VideoStim and DerivedVideoStim,
+    # but this may require reworking _input_type to handle disjunction rather
+    # than the current conjunction, or making the get_converter() code walk
+    # up the hierarchy and use superclass iterators.
+
+    _input_type = DerivedVideoStim
+    _output_type = ImageStim
+
+
+class ComplexTextIterator(StimCollectionIterator):
+
+    _input_type = ComplexTextStim
+    _output_type = TextStim

pliers/converters/multistep.py

@@ -31,7 +31,7 @@ def _convert(self, stim):
                     raise ValueError(msg)
             else:
                 converter = step
-            stim = converter.transform(stim)                    
+            stim = converter.transform(stim)
         return stim
 
 

pliers/converters/video.py

@@ -51,6 +51,9 @@ class FrameSamplingConverter(VideoToDerivedVideoConverter):
         top_n (int): takes top n frames sorted by the absolute difference
          with the next frame
     '''
+
+    _log_attributes = ('every', 'hertz', 'top_n')
+
     def __init__(self, every=None, hertz=None, top_n=None):
         super(FrameSamplingConverter, self).__init__()
         self.every = every
@@ -62,19 +65,12 @@ def _convert(self, video):
             frame_index = range(video.n_frames)
         else:
             frame_index = video.frame_index
-
-        if not hasattr(video, "history"):
-            history = pd.DataFrame(columns=["filter", "value", "n_frames"])
-        else:
-            history = video.history.copy()
 
         if self.every is not None:
             new_idx = range(video.n_frames)[::self.every]
-            history.loc[history.shape[0]]= ["every", self.every, len(new_idx)]
         elif self.hertz is not None:
             interval = int(video.fps / self.hertz)
             new_idx = range(video.n_frames)[::interval]
-            history.loc[history.shape[0]] = ["hertz", self.hertz, len(new_idx)]
         elif self.top_n is not None:
             import cv2
             diffs = []
@@ -85,7 +81,6 @@ def _convert(self, video):
                 diffs.append(sum(cv2.sumElems(cv2.absdiff(last, img))))
                 last = img
             new_idx = sorted(range(len(diffs)), key=lambda i: diffs[i], reverse=True)[:self.top_n]
-            history.loc[history.shape[0]] = ["top_n", self.top_n, len(new_idx)]
 
         frame_index = sorted(list(set(frame_index).intersection(new_idx)))
 
@@ -98,12 +93,9 @@ def _convert(self, video):
             else:
                 dur = (len(video.frames) / video.fps) - onsets[i]
 
-            elem = VideoFrameStim(video=video, frame_num=f,
-                                  duration=dur)
+            elem = VideoFrameStim(video=video, frame_num=f, duration=dur)
             elements.append(elem)
 
-        return DerivedVideoStim(filename=video.filename,
-                                elements=elements,
-                                frame_index=frame_index,
-                                history=history)
+        return DerivedVideoStim(filename=video.filename, elements=elements,
+                                frame_index=frame_index)
 

pliers/extractors/__init__.py

@@ -4,17 +4,28 @@
 import pandas as pd
 import numpy as np
 from collections import defaultdict
-
+from pliers import config
+from pliers.utils import memory
+from types import GeneratorType
 
 __all__ = ['api', 'audio', 'google', 'image', 'text', 'video']
 
 
 class Extractor(with_metaclass(ABCMeta, Transformer)):
     ''' Base class for Converters.'''
 
+    def __init__(self):
+        super(Extractor, self).__init__()
+        if config.cache_extractors:
+            self.transform = memory.cache(self.transform)
+
     def extract(self, stim, *args, **kwargs):
         return self.transform(stim, *args, **kwargs)
 
+    def transform(self, stim, *args, **kwargs):
+        result = super(Extractor, self).transform(stim, *args, **kwargs)
+        return list(result) if isinstance(result, GeneratorType) else result
+
     @abstractmethod
     def _extract(self, stim):
         pass
@@ -23,6 +34,7 @@ def _transform(self, stim, *args, **kwargs):
         return self._extract(stim, *args, **kwargs)
 
 
+
 class ExtractorResult(object):
 
     def __init__(self, data, stim, extractor, features=None, onsets=None,
@@ -31,6 +43,7 @@ def __init__(self, data, stim, extractor, features=None, onsets=None,
         self.stim = stim
         self.extractor = extractor
         self.features = features
+        self._history = None
         if onsets is None:
             onsets = stim.onset
         self.onsets = onsets if onsets is not None else np.nan
@@ -47,6 +60,14 @@ def to_df(self, stim_name=False):
             df.set_index('stim', append=True, inplace=True)
         return df
 
+    @property
+    def history(self):
+        return self._history
+
+    @history.setter
+    def history(self, history):
+        self._history = history
+
     @classmethod
     def merge_features(cls, results, extractor_names=True, stim_names=True):
         ''' Merge a list of ExtractorResults bound to the same Stim into a
@@ -61,7 +82,7 @@ def merge_features(cls, results, extractor_names=True, stim_names=True):
         '''
 
         # Make sure all ExtractorResults are associated with same Stim.
-        stims = set([r.stim for r in results])
+        stims = set([r.stim.name for r in results])
         dfs = [r.to_df() for r in results]
         if len(stims) > 1:
             raise ValueError("merge_features() can only be called on a set of "
@@ -94,16 +115,21 @@ def merge_features(cls, results, extractor_names=True, stim_names=True):
         if durations.apply(lambda x: x.nunique()<=1, axis=1).all():
             result = result.drop('duration', axis=1, level=1)
 
+        result.insert(0, 'class', results[0].stim.__class__.__name__)
+        result.insert(0, 'filename', results[0].stim.filename)
+        result.insert(0, 'history', str(results[0].history))
+
         if stim_names:
-            result['stim'] = list(stims)[0].name
+            result['stim'] = list(stims)[0]
             result.set_index('stim', append=True, inplace=True)
+            result = result.sort_index()
 
-        return result
+        return result.sort_values(['onset'])
 
     @classmethod
     def merge_stims(cls, results, stim_names=True):
         results = [r.to_df(True) if isinstance(r, ExtractorResult) else r for r in results]
-        return pd.concat(results, axis=0)
+        return pd.concat(results, axis=0).sort_values('onset')
 
 
 def merge_results(results, extractor_names=True, stim_names=True):

pliers/extractors/api.py

@@ -28,6 +28,8 @@ class IndicoAPIExtractor(ComplexTextExtractor):
         models (list): The names of the Indico models to use.  
     '''
 
+    _log_attributes = ('models',)
+
     def __init__(self, api_key=None, models=None):
         ComplexTextExtractor.__init__(self)
         if api_key is None:
@@ -92,6 +94,8 @@ class ClarifaiAPIExtractor(ImageExtractor):
             For example, ['food', 'animal'].
     '''
 
+    _log_attributes = ('model', 'select_classes')
+
     def __init__(self, app_id=None, app_secret=None, model=None, select_classes=None):
         ImageExtractor.__init__(self)
         if app_id is None or app_secret is None:
@@ -106,6 +110,8 @@ def __init__(self, app_id=None, app_secret=None, model=None, select_classes=None
         self.tagger = ClarifaiApi(app_id=app_id, app_secret=app_secret)
         if not (model is None):
             self.tagger.set_model(model)
+
+        self.model = model
 
         if select_classes is None:
             self.select_classes = None