--cca

Switch

--cca

Description

Performs CCA on the features and the outcomes and generates clusters (called components).

Argument and Default Value

Number of components to generate.

Details

This switch performs finds components that explain variance in both the features and the outcomes&controls, using the sparse CCA implementation in R. This first removes groups that have at least 4 non null values in the feature or outcome matrix, then performs softImpute (matrix completion) to get rid of the null values, and then performs CCA. Output will be in the form of (features x component) weights and (outcomes x component) weights, and the exact output format depends on the flags you specify (--rmatrix or --csv etc.).

Known issue — always crashes, even with R and rpy2 correctly installed. CCA.__init__() imports importr/ro/RNULLType from rpy2 as local variables inside the constructor — they're never stored on self or at module scope. Every other method that needs them (RsoftImpute(), called unconditionally by cca() via prepMatrices()) then references importr as if it were still in scope, hits a NameError, and that NameError is caught by an overly broad except NameError: warn("rpy2 cannot be imported"); sys.exit(1) — printing a misleading message that blames a missing dependency instead of the actual scoping bug. Verified: with real R (4.5.3) and rpy2 (3.5.2) both installed and independently confirmed importable, running --cca still prints rpy2 cannot be imported and exits with status 1 before performing any CCA computation. There is currently no way to run --cca successfully, regardless of environment.

Note that the number of components (the argument, sometimes called K) must satisfy:

There are a bunch of parameters for the sparse CCA function in R, the only ones that have a command line switch are penalties set on the "left" matrix (aka X, usually features) and the "right" matrix (aka Z, usually outcomes): --cca_penalty_feats, aka fwflag_cca_penaltyx --cca_penalty_outcomes, aka fwflag_cca_penaltyz To find which values to assign to these parameters, you can run --cca_permute If you want to do cca without using the features, use --cca_outcomes_vs_controls as an additional flag.

Other Switches

Required Switches:

• -d, -g, -t

• -f

• --outcome_table

• --outcomes

Optional Switches:

• --group_freq_thresh

• --outcome_controls

• --cca_penalty_feats

• --cca_penalty_outcomes

• --cca_outcomes_vs_controls

• --topic_tagcloud etc.

Example Commands

# Performs CCA on the topics and 15 diseases using 15 components
# Will output 2 HTML files called d0s0.K15.outcomes.html and d0s0.K15.feat.html
dlatkInterface.py -d county_disease -t messages_en -c cnty -f 'feat$cat_met_a30_2000_cp_w$messages_en$cnty$16to16' --group_freq_thresh 50000 --outcome_table topDeaths_comp_0910 --outcomes 01hea_aar 02mal_aar 03chr_aar 04cer_aar 05acc_aar 06alz_aar 07dia_aar 08nep_aar 09flu_aar 10sel_aar 11sep_aar 12liv_aar 13hyp_aar 14par_aar 15pne_aar --output_name ~/CCA/d0s0.K15 --rmatrix --cca 15

# Performs CCA on the topics and 15 diseases using 15 components, with penalties set to 0.5 and 0.5 for X and Z.
# Will output 2 HTML files called d0s0.K15.X0_5.Z0_5.outcomes.html and d0s0.K15.X0_5.Z0_5.feat.html, plus d0s0.K15.X0_5.Z0_5_topic_tagcloud.txt.
# It will also create the topic wordclouds in a separate directory.
dlatkInterface.py -d county_disease -t messages_en -c cnty -f 'feat$cat_met_a30_2000_cp_w$messages_en$cnty$16to16' --group_freq_thresh 50000 --outcome_table topDeaths_comp_0910 --outcomes 01hea_aar 02mal_aar 03chr_aar 04cer_aar 05acc_aar 06alz_aar 07dia_aar 08nep_aar 09flu_aar 10sel_aar 11sep_aar 12liv_aar 13hyp_aar 14par_aar 15pne_aar --output_name ~/CCA/d0s0.K15.X0_5.Z0_5 --rmatrix --cca 15 --topic_tagcloud --topic_lexicon met_a30_2000_freq_t50ll --cca_penaltyx .5 --cca_penaltyz .5 --csv --sort --make_topic_wordclouds