Skip to content

Latest commit

 

History

History
196 lines (173 loc) · 10.8 KB

Changelog.md

File metadata and controls

196 lines (173 loc) · 10.8 KB
Raw

Changelog

0.4.3 (November 23, 2022)

New extractors:

  • TFHubAudioExtractor, to support SPICE pitch extraction (#485)

Enhancements

  • Support non-English transcripts with RevAI (#484)

Fixes:

  • Fix CI testing

0.4.2 (October 20, 2022)

This is a minor release, with updates to the documentation, and various minor improvements:

  • Parameterize dtype for TFHubExtractor, add ImageResizingFilter + ImageRescalingFilter (#459)
  • Add Librosa fundamental frequency (#461)
  • Updated documentation (#445, #448, #447, #446), moved to ReadTheDocs (#482), and added video to README (#454)
  • Add more information to docstring for MFCCEnergyExtractor (#463)
  • Import ABC from collections.abc for Python 3.10 compatibility (#467)

The following extractor was deprecated:

  • Remove MicrosoftAPIFaceEmotionExtractor (#471)

Bug fixes and housekeeping:

  • FIX: Google APIs (remove discoveryServiceUrl) (#478)
  • FIX: Tests (#457, #477)
  • Move to Clarifai-GRPC to fix Clarifai extractors (#474)
  • FIX: Change excel engine (#455)
  • Build docker image with github actions (#452)
  • Update links from Naturalistic Data course (#458)

0.4.1 (February 26, 2021)

This is a minor release that adds one new extractor, and a resampling utility, as well as a large number of minor bugfixes.

New features:

  • Add TensorFlow hub extractor (#433)
  • Add ExtractedFeature resample utility (#430)

Bug fixes and other house keeping

  • Replace BERT with lighter extract in metric extractor tests
  • Fix pytest memory leak by running forked (#439)
  • Remove depracated pandas methods (#438)
  • Convert CI to Github Actions (#436)
  • Update HuggingFace BERT Extractors (#435)
  • Resample upsampling and add upsampling test case (#431)
  • Librosa links (#429)
  • Google API Regressions (#424)
  • Access string in numpy array (#422)
  • Update REAME to link to correct pliers docs page (#421)
  • Test regressions (#418)
  • Remove --no-user-group (#414)
  • Use multi-stage builds in Dockerfile to reduce final image size (#411)
  • Remove instance of Python2 super (#410)
  • Update librosa feature links (#409)
  • If Google Face Extractor result landmark has no type, skip. (#406)
  • librosa import error (numba) (#405)
  • Remove soundfile as dependency (#408)

0.4.0 (July 14, 2020)

This is a major release that adds support for several new feature extraction services/models, as well as a large number of minor bugfixes and improvements.

New features and major improvements/bugfixes:

  • Add WordCounterExtractor (#364)
  • Add PretrainedBertEncodingExtractor from transformers #365
  • Add lemmatization to stemming filter (WordStemmingFilter) #366
  • Add Lancaster Sensorimotor Norms to PredefinedDictionaryExtractor #367
  • Remove deprecated Indico API extractors #369
  • Add AudioResamplingFilter #374
  • Add AudiosetLabelExtractor from Yamnet architecture #379 #399
  • Add BertLMExtractor, BertSentimentExtractor, BertSequenceEncodingExtractor #383 #397
  • Add log attributes to to_df #389
  • Add MetricExtractor #390 #398
  • Add ExtractorResult --> SeriesStim converter #394

Minor improvements and bug fixes:

  • Fixes case of missing landmark type in GoogleVisionAPIFaceExtractor (#406)
  • Removed soundfile as dependency (#408)
  • Fix librosa import erroor (#405)
  • Fixes to Clarifai face extractor (#335, #357)
  • Added optional dependencies as 'extras' for pip install #358
  • Patch operations request for Google Video Intelligence #360
  • Miscellaneous fixes to tests and Travis CI #372 #375 #378 #380 #381
  • Fix index_cols behavior in to_df function #376 #388
  • Drop vestigial Python 2 code #392
  • Upgrade Python syntax and add python_requires for 3.5+ #393

0.3.0 (June 8, 2019)

This is a major release that adds support for several new feature extraction services/models, as well as a large number of minor bugfixes and improvements.

New features and major improvements/bugfixes:

  • Basic support for Google Natural Language API (#306)
  • Basic support for Google Video Intelligence API (#288)
  • Support for Rev.ai speech-to-text converter (#334)
  • Added several librosa features (#320)
  • Support for pre-trained Keras-based image classifiers (#341, #344)
  • Added .srt output functionality to ComplexTextStim (#301)
  • Added a get_bytestring() method to ImageStim and VideoStim classes (#324)
  • Expanded/improved PredefinedDictionaryExtractor list (#339, #351)
  • Adds ImageResizeFilter (#342)
  • Improved/fixed Dockerfile (#343)

Minor improvements and bug fixes:

  • Fixed bug in GoogleSpeechAPIConverter that overwrote all but last speech block (#316)
  • Improved sampling accuracy in FrameSamplingFilter (#301)
  • Fixed bug in RGB information returned by GoogleVisionAPI (#296)
  • Fixed caching of transformations on file paths (#286)
  • Adds support for a few config options in Google Video Intelligence extractors (#337)
  • Various other minor improvements and fixes (#301, #308, #312, #324, #332)

0.2.3 (April 7, 2018)

This is a minor release that adds several new features and bug fixes:

  • A number improvements to API transformers (custom marker for unit tests; key validation; large job limits; etc.; see #270)
  • Enhance API transformers to work from remote URLs if available, without requiring local file download (#282)
  • Protected timing variable names to prevent collisions in to_df calls (#281)
  • Add caching to transformers that inherit from BatchTransformerMixin (#283)
  • Fixes to update utility (#273)
  • Updated docstrings (#282)

0.2.2 (March 1, 2018)

This is a minor release that adds several new features and bug fixes:

  • Support for several Microsoft Vision services, including the Face API and Vision API (#259)
  • Improvements to the Graph API (#254):
    • Graphs can now return Stims in cases where terminal nodes are Converters or Filters
    • Serialization and de-serialization from JSON
    • A helper add_chain method that simplifies construction of completely linear Graphs
    • Improved graph plotting, including representation of node types using color and line type
  • Improved internal handling and propagation of temporal properties (onset, duration, and order; #261)
  • Refactored the internal handling/formatting of raw feature extraction results (#261)
  • Added temporal cropping filters that make it easy to crop audio and video clips to specified boundaries (#244)
  • Reorganization of the API Transformer hierarchy (#266)
  • Minor improvements to to text filters (e.g., lower-casing and tokenization; #246)
  • Added basic support for duecredit (thanks to @yarikoptic; #254)
  • Several minor bug fixes

0.2.1 (January 31, 2018)

This is a bugfix release that addresses an installation bug on bare environments (thanks to @mgxd).

0.2 (January 29, 2018)

This is a major release that adds many new features and transformers. Changes include:

  • Sphinx documentation!
  • New package features
    • An updated extractor result management API with simplified merging mechanisms (see #232 for more details)
    • Ability to create transformer Graphs via a JSON specification
    • Scitkit-learn integration: ability to use pliers Graphs or Extractors as sklearn Transformers that can be plugged into full Pipelines
    • Batch transformation: added ability to transform stimuli in batches when the transformer inherits BatchTransformerMixin, but a batch size must be given
    • Static versioning for transformers (0.1 indicates initial working transformer, 1.0+ indicates fully-tested and often used transformer)
    • Added TweetStim, that can be initialized using Twitter's API, to the stimuli hierarchy
    • A update checker utility (in pliers.utils) that allows users to track changes to the output of different transformers for user-provided stimuli or a hand-selected battery of stimuli from the pliers test module
    • A revamped and improved config module with exposed accessor methods
  • New Transformers:
    • NLP Transformers:
      • TextVectorizerExtractor uses sklearn Vectorizers to extract bag-of-X features from a batch of text documents
      • WordEmbeddingExtractor uses gensim to extract word embeddings using a user-provided word2vec file
      • VADERSentimentExtractor uses nltk's VADER sentiment analysis tool to do sentiment analysis (w/o needing an API)
      • TokenizingFilter that splits a TextStim into several TextStims with one token in each
      • TokenRemovalFilter that removes specified tokens (e.g. stopwords, punctuation) from a TextStim
    • Audio feature extraction using librosa, wrapping all of the feature extraction methods https://librosa.github.io/librosa/feature.html
    • Image Filters
      • ImageCroppingFilter crops an image using a user-specified box (or just trims off black borders by default)
      • PillowImageFilter uses common simple filters from PIL to filter an image. Examples: edge, blur, contour, min filters
    • Google Vision API image web entity extraction
    • Face recognition extractor using the face_recognition package
  • Fixes or enhancements to existing features:
    • Better identifying of stimulus uniqueness (for hashing purposes), to prevent strange merges when using merge_results
    • Temporary file management for transformers that require on-disk file names
    • Better support for exporting to vertically long Pandas dataframes
    • Added ability to flatten the multi-index returned from merge_results
    • Using v2 of the Clarifai API
    • Added ability to control whether or not metadata is presented in merged or single ExtractorResults
    • Removed DerivedVideoStim class and replaced with VideoFrameCollectionStim to clean up VideoStim hierarchy
    • Standardized dependency handling with attempt_to_import function calls
    • Proper calculation of AudioStim sampling_rate
    • Added ability to use the GoogleSpeechAPIConverter to extract word onsets and durations as well
  • A number of other minor bug fixes and cleaning of the codebase

0.1.1 (March 9, 2017)

This is a minor release that adds a number of new (minor) features and a bunch of bug fixes. Changes include:

  • Added very basic parallelization support
  • Added support for generating graph diagrams using PyGraphViz (via Graph.plot())
  • Added a new RemoteStim that represents remote, uninitialized resources
  • Added a new WordStemmingFilter that stems text using nltk
  • Added support for Google Cloud Vision's safe image detection
  • IBMSpeechAPIConverter now takes an argument indicating whether to return words or phrases
  • Added support for image-based Indico API models
  • Add support for multiple results returned by IBM speech converter
  • General clean-up, code styling, and docstring addition
  • Bug fixes:
    • Update various Transformers to reflect changes to APIs
    • Fixed dataset import bug that prevented PyPI-installed 0.1 from importing properly
    • Update args to numpy functions to ensure numpy 1.12+ compatibility
    • Various other minor bug fixes

0.1 (January 17, 2017)

First major release. Nothing before this counts.