.. _changelog: ========= Changelog ========= 1.2.0 (2021-03-11) ------------------ - New feature: Huggingface interface with --add_embedding (Bert, Roberta, XLNet, GPT2, etc.) - New feature: pymallet interface for creating LDA topics - New feature: optional sqlite backend - New feature: MySQL configuration read from config files (~/.my.cnf) instead of command line - Improving feature: dropped the 16to16 notation for default encoding (i.e., non-transformed) feature tables - Improving feature: new notation for group flag (--group or -g), in addition to previous version (--correl_field or -c) - Improving feature: default for running n-fold cross validation is to use all controls, use --all_control_combinations for all combinations - Improving feature: additional classification metrics printed to terminal - Improving feature: support for chinese segmentation in mallet - Improving feature: normalize lexicon scores by words with category (rather than all words), --lexicon_weighting - Improving feature: all external tools moved to ~/dlatk_tools - Bug: fixed bugs in --regression_to_lex 1.1.6 (2019-07-31) ------------------ - New feature: bert feature extraction v1 ready - New feature: added --keep_low_variance_outcomes - New feature: added --outliers_to_mean to classifyPredictor - New feature: added --cohens_d flag - New feature: added factor adaptation code - New feature: added --multiclass flag - New feature: added new flag --predict_probabilities_to_feats - Improving feature: added effect size ranges to ngram wordcloud filenames - Improving feature: new default is to *not* print duplicate topic wordclouds, use --keep_duplicates to turn this off - Improving feature: beta classification auc ensembling - Improving feature: add _intercept=1 to all groups in lex feature tables - Improving feature: --classification_to_lexicon works with topic level features and multiple feature tables - Improving feature: remove underscores when uploading multiword lexica, use --keep_underscores to turn off - Improving feature: added mysql methods for checking indices - Improving feature: docker documentation - Improving feature: cleaned up regression predictor output - Improving feature: using default random seed throughout classes - Improving feature: fixed bug in interactions and language filtering - Improving feature: added printing of true values to pred_csv - Improving feature: teal colorscheme, aliases for 'tagcloud' and 'wordcloud' - Improving feature: metric names in wordcloud output - Bug: fixed n_iter shufflesplit outdated param - Bug: fixed nan pvalue issue in control corrections - Bug: fixed --log transform table naming convention - Bug: fixed bug in dlatk/classifyPredictor.py, scaler fit_transform was called during _multiXpredict, changed to transform - Bug: correct version of wordcloud python module is 1.1.3 1.1.5 (2018-03-02) ------------------ - Improving feature: changes to LDA process: topicExtractor now accessible inside dlatkInterface - Improving feature: --lex_interface: lexInterface flags accessible inside dlatkInterface - Improving feature: changes default lexicon based feature table names - Improving feature: --categorical flag removes one outcome for any column with only two values. - Improving feature: mysqlMethods changed to mysqlmethods - Improving feature: default lexicon MySQL database now called "dlatk_lexica" (old database called "permaLexicon") - Bug fix: updated ridgehighcv to fix integer bug in sklearn 1.1.4 (2018-01-11) ------------------ - Bug fix: added __init__.py file to Tools directory to fix pip install 1.1.3 (2018-01-09) ------------------ - Improving feature: created the alias --show_feat_tables for --ls 1.1.2 (2018-01-09) ------------------ - New Feature: added --show_tables - New Feature: added --describe_tables - New Feature: added --create_random_sample 1.1.1 (2017-11-30) ------------------ - New Feature: added --categories_to_binary - New Feature: added --create_collocation_scores - New Feature: added --weighted_sample for running weight regression - New Feature: added --reduced_lexicon flag for creating super topics - New Feature: added --extension - Improving feature: allow LDA messages to have non-numeric message id - Improving feature: removed extra '.' from tagcloud filenames - Improving feature: moved location of Stanford Segmenter to dlaConstants - Bug fix: in --predict_regression_to_outcome_table and --predict_regression_to_feats - Bug fix: in setup.py for adding Tools directory - Bug fix: fixed bug in super topics (added auto increment id) 1.1.0 (2017-06-29) ------------------ - New feature: added --n_components / --num_factors flag to --fit_reducer - New feature: --clean_cloud flag working for n_gram correlation, all clouds, and topic clouds - New feature: added p correction to correlate --auc - New feature: messageAnnotator and messageTransformer classes added - New feature: clustering.py changed to dimensionReducer.py - New feature: FeatureWorker changed to DLAWorker - New feature: --clean_messages flag for anonymizing message tables - Improving feature: residualized control model in --combo_test_regression - Improving feature: changed f1 to macro f1 instead of pos-class f1 - Bug fix: in --predict_regression_to_outcome_table - Bug fix: in --regression_to_lexicon 1.0.1 (2016-11-21) ------------------ - New feature: --deduplicate flag added to remove duplicate tweets within user - New feature: --spam_filter added - New feature: --cleanmessages added to --deduplicate and --spam_filter - New feature: --fold_column added to classifyPredictor and regressionPredictor - Bug fix: --print_csv - Bug fix: csv from binary to text mode lexInterface - Bug fix: add_postimexdiff, nlp server - Improving feature: error messages when incorrectly specifying --p_correction - Improving feature: changed _tok message tables to always be stored as longtext 1.0.0 (2016-10-21) ------------------ - New feature: python 3.5 version, changed from FeatureWorker to dlatk - New feature: --densify_table - New feature: --language_filter and --clean_messages - New feature: --make_all_topic_wordclouds - New feature: --no_lower flag added for --add_char_ngrams - New feature: confidence intervals added to --correlate, --rmatrix and --csv - New feature: --add_message_id flag replaces LexicaInterface/addMessageID.py - New feature: --ls - New feature: stratified classification - New feature: --where - New feature: --stratify_folds added to --combo_test_classifiers - Bug fix: using multiple feature tables with --to_file - Improving feature: batch insert in tfidf - Improving feature: make_wordclouds/make_topic_wordcloud error catch - Improving feature: removed lda.py 0.6.1 (2016-07-05) ------------------ - New feature: install notes for OSX - Bug fix: using multiple feature tables with --to_file - Improving feature: Default MySQL host changed from 'localhost' to 127.0.0.1 0.6.0 (2016-06-15) ------------------ - Bug fix: in makeBlackWhiteList - Bug fix: in fiftyChecks, added random seed - Bug fix: for using old pickle files - Improving feature: added unicode try/except in wordcloud module and print csv - Improving feature: removed hardcoded utf8 and hardcoded table charset - New feature: --feat_selection_string - New feature: --add_corp_lex_table - New feature: new models added to regressionPredictor (ridgefirstpasscv, ridgehighcv, ridgelowcv) - New feature: --add_char_ngrams - New feature: --no_outcomes and --no_controls 0.5.0 (2016-04-01) ------------------ 0.4.0 (2015-08-17) ------------------