diff options
author | Alexander_Kabui | 2022-08-16 12:11:36 +0300 |
---|---|---|
committer | Alexander_Kabui | 2022-08-16 12:11:36 +0300 |
commit | fa8ef3e466e3919648e1d4cf9c38ed30328fc7a6 (patch) | |
tree | ac30a329e9c0b7a88010ec11a3195ed142fa3050 /wqflask | |
parent | cda1370d5712ae3c756215ef848dedc99cd5504d (diff) | |
download | genenetwork2-fa8ef3e466e3919648e1d4cf9c38ed30328fc7a6.tar.gz |
minor fixes for computing all correlations
Diffstat (limited to 'wqflask')
-rw-r--r-- | wqflask/wqflask/correlation/rust_correlation.py | 69 |
1 files changed, 38 insertions, 31 deletions
diff --git a/wqflask/wqflask/correlation/rust_correlation.py b/wqflask/wqflask/correlation/rust_correlation.py index 94720f54..2a2ad4a0 100644 --- a/wqflask/wqflask/correlation/rust_correlation.py +++ b/wqflask/wqflask/correlation/rust_correlation.py @@ -39,15 +39,14 @@ def chunk_dataset(dataset,steps,name): strains = [trait_name] + [str(value) for (trait_name, strain, value) in matrix] results.append(",".join(strains)) - breakpoint() return results def compute_top_n_sample(start_vars, dataset, trait_list): - """only if dataset is of type probeset""" - - + """check if dataset is of type probeset""" + if dataset.type!= "Probeset": + return {} def __fetch_sample_ids__(samples_vals, samples_group): @@ -73,19 +72,9 @@ def compute_top_n_sample(start_vars, dataset, trait_list): ) - return dict(curr.fetchall()) - - - - - + return (sample_data,dict(curr.fetchall())) - - - - - - ty = __fetch_sample_ids__(start_vars["sample_vals"], start_vars["corr_samples_group"]) + (sample_data,sample_ids) = __fetch_sample_ids__(start_vars["sample_vals"], start_vars["corr_samples_group"]) @@ -93,6 +82,8 @@ def compute_top_n_sample(start_vars, dataset, trait_list): curr = conn.cursor() + #fetching strain data in bulk + curr.execute( """ @@ -104,15 +95,14 @@ def compute_top_n_sample(start_vars, dataset, trait_list): and ProbeSetFreeze.Name = '{}' and ProbeSet.Name in {} and ProbeSet.Id = ProbeSetXRef.ProbeSetId) - """.format(create_in_clause(list(ty.values())),dataset.name,create_in_clause(trait_list)) + """.format(create_in_clause(list(sample_ids.values())),dataset.name,create_in_clause(trait_list)) ) + corr_data = chunk_dataset(list(curr.fetchall()),len(sample_ids.values()),dataset.name) - - - return chunk_dataset(list(curr.fetchall()),len(ty.values()),dataset.name) + return run_correlation(corr_data,list(sample_data.values()),"pearson",",") def compute_top_n_lit(corr_results, this_dataset, this_trait) -> dict: @@ -170,7 +160,10 @@ def merge_results(dict_a: dict, dict_b: dict, dict_c: dict) -> list[dict]: **dict_c.get(trait_name, {}) } } - return [__merge__(tname, tcorrs) for tname, tcorrs in dict_a.items()] + results = [__merge__(tname, tcorrs) for tname, tcorrs in dict_a.items()] + + + return results def __compute_sample_corr__( @@ -249,27 +242,41 @@ def compute_correlation_rust( } results = corr_type_fns[corr_type]( start_vars, corr_type, method, n_top, target_trait_info) + # END: Replace this with `match ...` once we hit Python 3.10 - top_tissue_results = {} - top_lit_results = {} + top_a = top_b = {} - results = compute_top_n_sample(start_vars,target_dataset,list(results.keys())) + if compute_all: + if corr_type == "sample": + top_a = compute_top_n_tissue( + this_dataset, this_trait, results, method) + + top_b = compute_top_n_lit(results, this_dataset, this_trait) - breakpoint() - if compute_all: - # example compute of compute both correlation - top_tissue_results = compute_top_n_tissue( + elif corr_type == "lit": + + #currently fails for lit + + top_a = compute_top_n_sample(start_vars,target_dataset,list(results.keys())) + top_b = compute_top_n_tissue( this_dataset, this_trait, results, method) - top_lit_results = compute_top_n_lit(results, this_dataset, this_trait) - return { + else: + + top_a = compute_top_n_sample(start_vars,target_dataset,list(results.keys())) + + top_b = compute_top_n_lit(results, this_dataset, this_trait) + + + + return { "correlation_results": merge_results( - results, top_tissue_results, top_lit_results), + results, top_a, top_b), "this_trait": this_trait.name, "target_dataset": start_vars['corr_dataset'], "return_results": n_top |