diff options
-rw-r--r-- | wqflask/wqflask/correlation/show_corr_results.py | 46 |
1 files changed, 44 insertions, 2 deletions
diff --git a/wqflask/wqflask/correlation/show_corr_results.py b/wqflask/wqflask/correlation/show_corr_results.py index ab8781cb..bc3b4fd7 100644 --- a/wqflask/wqflask/correlation/show_corr_results.py +++ b/wqflask/wqflask/correlation/show_corr_results.py @@ -73,6 +73,49 @@ def set_template_vars(start_vars, correlation_data): return correlation_data + min_expr = get_float(start_vars, 'min_expr') + p_range_lower = get_float(start_vars, 'p_range_lower', -1.0) + p_range_upper = get_float(start_vars, 'p_range_upper', 1.0) + + if ('loc_chr' in start_vars and + 'min_loc_mb' in start_vars and + 'max_loc_mb' in start_vars): + + location_chr = get_string(start_vars, 'loc_chr') + min_location_mb = get_int(start_vars, 'min_loc_mb') + max_location_mb = get_int(start_vars, 'max_loc_mb') + else: + location_chr = min_location_mb = max_location_mb = None + + +def get_user_filters(start_vars): + """ a user can filter the results based on different criterias""" + (min_expr, p_min, p_max) = ( + get_float(start_vars, 'min_expr'), + get_float(start_vars, 'p_range_lower', -1.0), + get_float(start_vars, 'p_range_upper', 1.0) + ) + + if ["loc_chr", "min_loc_mb", "max_location_mb"] in start_vars: + + location_chr = get_string(start_vars, "loc_chr") + min_location_mb = get_int(start_vars, "min_loc_mb") + max_location_mb = get_int(start_vars, "max_loc_mb") + + else: + location_chr = min_location_mb = max_location_mb = None + + return { + + "min_expr": min_expr, + "p_range_lower": p_min, + "p_range_upper": p_max, + "location_chr": location_chr, + "min_location_mb": min_location_mb, + "max_location_mb": max_location_mb + + } + def generate_table_metadata(all_traits, dataset_metadata, dataset_obj): @@ -87,7 +130,7 @@ def generate_table_metadata(all_traits, dataset_metadata, dataset_obj): return (dataset_metadata | ({trait["name"]: trait for trait in metadata})) -def populate_table(dataset_metadata,target_dataset,corr_results): +def populate_table(dataset_metadata, target_dataset, corr_results): def __populate_trait__(idx, target_trait, target_dataset): results_dict = {} results_dict['index'] = idx + 1 # @@ -228,7 +271,6 @@ def correlation_json_for_table(start_vars, correlation_data, this_trait, this_da else: continue - results_dict = {} results_dict['index'] = i + 1 results_dict['trait_id'] = target_trait['name'] |