From 361157f992af320abf73d93d9143b25017068184 Mon Sep 17 00:00:00 2001 From: Alexander_Kabui Date: Wed, 10 Aug 2022 08:43:11 +0300 Subject: code to merge results;minor fixes --- wqflask/wqflask/correlation/rust_correlation.py | 51 +++++++++++++++++++++---- 1 file changed, 44 insertions(+), 7 deletions(-) (limited to 'wqflask') diff --git a/wqflask/wqflask/correlation/rust_correlation.py b/wqflask/wqflask/correlation/rust_correlation.py index 161215c5..8431f179 100644 --- a/wqflask/wqflask/correlation/rust_correlation.py +++ b/wqflask/wqflask/correlation/rust_correlation.py @@ -11,8 +11,6 @@ from gn3.computations.rust_correlation import parse_tissue_corr_data from gn3.db_utils import database_connector - - def compute_top_n_lit(corr_results, this_dataset, this_trait): (this_trait_geneid, geneid_dict, species) = do_lit_correlation( this_trait, this_dataset) @@ -31,9 +29,10 @@ def compute_top_n_lit(corr_results, this_dataset, this_trait): return correlation_results - def compute_top_n_tissue(this_dataset, this_trait, traits, method): + # refactor lots of rpt + trait_symbol_dict = dict({trait_name: symbol for ( trait_name, symbol) in this_dataset.retrieve_genes("Symbol").items() if traits.get(trait_name)}) @@ -48,11 +47,32 @@ def compute_top_n_tissue(this_dataset, this_trait, traits, method): if data: return run_correlation( - data[1], data[0], method, ",","tissue") + data[1], data[0], method, ",", "tissue") return {} +def merge_results(dict_a, dict_b, dict_c): + """code to merge diff corr into individual dicts + a""" + + correlation_results = [] + + for (key, val) in dict_a.items(): + + if key in dict_b: + + dict_a[key].update(dict_b[key]) + + if key in dict_c: + + dict_a[key].update(dict_c[key]) + + correlation_results.append({key: dict_a[key]}) + + return correlation_results + + def compute_correlation_rust(start_vars: dict, corr_type: str, method: str = "pearson", n_top: int = 500): """function to compute correlation""" @@ -75,10 +95,27 @@ def compute_correlation_rust(start_vars: dict, corr_type: str, r = ",".join(lts) target_data.append(r) - results = run_correlation(target_data, ",".join( - [str(x) for x in list(sample_data.values())]), method, ",") + results = run_correlation(target_data, + list(sample_data.values()), + method, + ",", + corr_type, + n_top) + + # example compute of compute both correlation + + + + top_tissue_results = compute_top_n_tissue(this_dataset,this_trait,results,method) + + + top_lit_results = compute_top_n_lit(results,this_dataset,this_trait) + + + # merging the results + results = merge_results(results,top_tissue_results,top_lit_results) if corr_type == "tissue": @@ -95,7 +132,7 @@ def compute_correlation_rust(start_vars: dict, corr_type: str, if data: results = run_correlation( - data[1], data[0], method, ",","tissue") + data[1], data[0], method, ",", "tissue") return {"correlation_results": results, "this_trait": this_trait.name, -- cgit v1.2.3