Merge fef368d into 7b65ddf

.travis.yml

@@ -1,19 +1,18 @@
 sudo: required
 dist: trusty
-env:
-  global:
-  - CONDA_DEPS="pip pytest numpy pillow pandas requests scipy nltk six tqdm seaborn opencv
-    matplotlib pathos tensorflow contextlib2"
+language: python
 matrix:
   include:
   - os: linux
     env:
     - PYTHON_VERSION=3.5
     - MINICONDA_URL="https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh"
+    python: 3.5
   - os: linux
     env:
-    - PYTHON_VERSION=2.7
-    - MINICONDA_URL="https://repo.continuum.io/miniconda/Miniconda2-latest-Linux-x86_64.sh"
+    - PYTHON_VERSION=3.6
+    - MINICONDA_URL="https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh"
+    python: 3.6
 before_install:
 - openssl aes-256-cbc -K $encrypted_a0a62c26415d_key -iv $encrypted_a0a62c26415d_iv
   -in pliers/tests/credentials/google.json.enc -out pliers/tests/credentials/google.json -d || true
@@ -30,24 +29,25 @@ before_install:
 - conda config --add channels conda-forge
 - conda update -y conda
 - rm -rf /home/travis/miniconda/envs/test-env
-- conda create -q -n test-env python=$PYTHON_VERSION $CONDA_DEPS
+- conda create -q -n test-env python=$PYTHON_VERSION
 - source activate test-env
 install:
 - sudo apt-get install libboost-python-dev
 - pip install --upgrade --ignore-installed setuptools
-- pip install python-magic moviepy coveralls pytest-cov pysrt xlrd pytesseract
-- pip install clarifai==2.1.0
-- pip install SpeechRecognition IndicoIo pygraphviz sklearn python-twitter gensim
-  google-compute-engine librosa face_recognition google-api-python-client
-- pip install spacy
+- pip install -r requirements.txt --upgrade
+- pip install -r optional-dependencies.txt --upgrade 
+- pip install --upgrade coveralls pytest-cov
+
 before_script:
 - python -m pliers.support.download
+- python -m spacy download en_core_web_sm
 script:
-- py.test --pyargs pliers --cov-report term-missing --cov=pliers
+- py.test --pyargs pliers --cov-report term-missing --cov=pliers -m "not requires_payment"
 after_success:
 - coveralls
 before_cache:
 - conda clean --tarballs --packages --index-cache
 cache:
+  pip: true
   directories:
   - "$HOME/miniconda"

README.md

@@ -1,6 +1,6 @@
 # pliers
 
-A Python 2/3 package for automated feature extraction.
+A Python 3 package for automated feature extraction.
 
 ## Status
 * [![Build Status](https://travis-ci.org/tyarkoni/pliers.svg?branch=master)](https://travis-ci.org/tyarkoni/pliers)

docs/stimuli.rst

@@ -65,7 +65,7 @@ Consider this code:
     tac = TranscribedAudioCompoundStim(audio, text)
 
     # Construct a Graph with two extractors
-    graph = Graph(['PartOfSpeechExtractor', 'RMSEExtractor'])
+    graph = Graph(['PartOfSpeechExtractor', 'RMSExtractor'])
 
     # Apply the extractors to the Stim
     result = graph.transform(tac)

docs/transformers.rst

@@ -52,7 +52,7 @@ At present, pliers implements several dozen |Extractor| classes that span a wide
   MelspectrogramExtractor
   MFCCExtractor
   PolyFeaturesExtractor
-  RMSEExtractor
+  RMSExtractor
   SpectralCentroidExtractor
   SpectralBandwidthExtractor
   SpectralContrastExtractor

optional-dependencies.txt

@@ -4,8 +4,9 @@ face_recognition
 python-twitter
 gensim
 google-api-python-client
+google-compute-engine
 IndicoIo
-librosa
+librosa>=0.6.3
 matplotlib
 opencv-python
 pathos
@@ -15,5 +16,7 @@ pytesseract
 python-twitter
 scikit-learn
 seaborn
+spacy
 SpeechRecognition>=3.6.0
 tensorflow>=1.0.0
+xlrd

pliers/config.py

@@ -12,7 +12,8 @@
 
 _default_converters = {
     'AudioStim->TextStim':
-        ('IBMSpeechAPIConverter', 'WitTranscriptionConverter'),
+        ('GoogleSpeechAPIConverter', 'IBMSpeechAPIConverter',
+         'WitTranscriptionConverter'),
     'ImageStim->TextStim':
         ('GoogleVisionAPITextConverter', 'TesseractConverter')
 }

pliers/converters/api/ibm.py

@@ -8,9 +8,9 @@
 from pliers.utils import attempt_to_import, verify_dependencies
 from pliers.converters.audio import AudioToTextConverter
 from pliers.transformers.api import APITransformer
-from six.moves.urllib.parse import urlencode
-from six.moves.urllib.request import Request, urlopen
-from six.moves.urllib.error import URLError, HTTPError
+from urllib.parse import urlencode
+from urllib.request import Request, urlopen
+from urllib.error import URLError, HTTPError
 
 sr = attempt_to_import('speech_recognition', 'sr')
 
@@ -33,14 +33,16 @@ class IBMSpeechAPIConverter(APITransformer, AudioToTextConverter):
             supported.
         rate_limit (int): The minimum number of seconds required between
             transform calls on this Transformer.
+        model (str): The model to use for speech recognition (e.g., 'en-US',
+            'zh-CN', etc.). Don't include the "_BroadbandModel" suffix. 
     '''
 
     _env_keys = ('IBM_USERNAME', 'IBM_PASSWORD')
-    _log_attributes = ('username', 'password', 'resolution')
+    _log_attributes = ('username', 'password', 'resolution', 'model')
     VERSION = '1.0'
 
     def __init__(self, username=None, password=None, resolution='words',
-                 rate_limit=None):
+                 rate_limit=None, model='en-US'):
         verify_dependencies(['sr'])
         if username is None or password is None:
             try:
@@ -53,6 +55,7 @@ def __init__(self, username=None, password=None, resolution='words',
         self.username = username
         self.password = password
         self.resolution = resolution
+        self.model = model
         super(IBMSpeechAPIConverter, self).__init__(rate_limit=rate_limit)
 
     @property
@@ -119,7 +122,7 @@ def _query_api(self, clip):
             convert_rate=None if clip.sample_rate >= 16000 else 16000,
             convert_width=None if clip.sample_width >= 2 else 2
         )
-        model = "{0}_BroadbandModel".format("en-US")
+        model = "{0}_BroadbandModel".format(self.model)
         url = "https://stream.watsonplatform.net/speech-to-text/api/v1/recognize?{0}".format(urlencode({
             "profanity_filter": "false",
             "continuous": "true",

pliers/extractors/__init__.py

@@ -39,7 +39,6 @@
                     SpectralContrastExtractor,
                     SpectralRolloffExtractor,
                     PolyFeaturesExtractor,
-                    RMSEExtractor,
                     ZeroCrossingRateExtractor,
                     ChromaSTFTExtractor,
                     ChromaCQTExtractor,
@@ -83,7 +82,6 @@
     'SpectralContrastExtractor',
     'SpectralRolloffExtractor',
     'PolyFeaturesExtractor',
-    'RMSEExtractor',
     'ZeroCrossingRateExtractor',
     'ChromaSTFTExtractor',
     'ChromaCQTExtractor',

pliers/extractors/api/clarifai.py

@@ -182,7 +182,15 @@ def _to_df(self, result):
             else:
                 end = frames[i+1]['frame_info']['time'] / 1000.0
             onsets.append(onset)
-            durations.append(end - onset)
+            # NOTE: As of Clarifai API v2 and client library 2.6.1, the API
+            # returns more frames than it should—at least for some videos.
+            # E.g., given a 5.5 second clip, it may return 7 frames, with the
+            # last beginning at 6000 ms. Since this appears to be a problem on
+            # the Clarifai end, and it's not actually clear how they're getting
+            # this imaginary frame (I'm guessing it's the very last frame?),
+            # we're not going to do anything about it here, except to make sure
+            # that durations aren't negative.
+            durations.append(max([end - onset, 0]))
 
         result._onsets = onsets
         result._durations = durations

pliers/extractors/api/google.py

@@ -1,6 +1,5 @@
 ''' Google API-based feature extraction classes. '''
 
-import base64
 from pliers.extractors.image import ImageExtractor
 from pliers.extractors.text import TextExtractor
 from pliers.extractors.video import VideoExtractor
@@ -13,6 +12,8 @@
 import pandas as pd
 import logging
 import time
+import warnings
+import os
 from collections import defaultdict
 
 
@@ -208,16 +209,21 @@ def _query_operations(self, name):
         return request_obj.execute(num_retries=self.num_retries)
 
     def _build_request(self, stim):
-        with stim.get_filename() as filename:
-            with open(filename, 'rb') as f:
-                vid_data = f.read()
 
         context = self.config if self.config else {}
         if self.segments:
             context['segments'] = self.segments
 
+        with stim.get_filename() as filename:
+            size = os.path.getsize(filename)
+            LIMIT = 524288000
+            if size > LIMIT:
+                warnings.warn("Video file is very large ({} bytes) and may "
+                              "exceed the Google Video Intelligence payload "
+                              "limit ({} bytes).".format(size, LIMIT))
+
         request = {
-            'inputContent': base64.b64encode(vid_data).decode(),
+            'inputContent': stim.get_bytestring(),
             'features': self.features,
             'videoContext': context
         }

pliers/extractors/api/indico.py

@@ -11,6 +11,7 @@
 from pliers.transformers.api import APITransformer
 from pliers.utils import attempt_to_import, verify_dependencies
 import pandas as pd
+import numpy as np
 
 indicoio = attempt_to_import('indicoio')
 
@@ -84,6 +85,7 @@ def _get_tokens(self, stims):
         return [stim.data for stim in stims if stim.data is not None]
 
     def _extract(self, stims):
+        stims = list(stims)
         tokens = self._get_tokens(stims)
         scores = [model(tokens) for model in self.models]
 
@@ -157,5 +159,7 @@ def _get_tokens(self, stims):
             if s.url:
                 toks.append(s.url)
             elif s.data is not None:
-                toks.append(s.data)
+                # IndicoIO breaks if given subclasses of ndarray, and data is
+                # an imageio Image instance, so we explicitly convert.
+                toks.append(np.array(s.data))
         return toks