From 05740a60d6616f28751f96ca30adeb524f4369ad Mon Sep 17 00:00:00 2001
From: Frederick Muriuki Muriithi
Date: Mon, 1 Nov 2021 10:49:35 +0300
Subject: Implement `compute_partial_correlations_fast`

Issue:
https://github.com/genenetwork/gn-gemtext-threads/blob/main/topics/gn1-migration-to-gn2/partial-correlations.gmi

* Implement `compute_partial_correlations_fast` that is a partial migration of
  `web.webqtl.correlation.PartialCorrDBPage.getPartialCorrelationsFast` in
  GN1.

  This function will probably be reworked once the dependencies are fully
  migrated.

  It also needs tests to be added.
---
 gn3/computations/partial_correlations.py | 49 ++++++++++++++++++++++++++++++++
 1 file changed, 49 insertions(+)

(limited to 'gn3/computations')

diff --git a/gn3/computations/partial_correlations.py b/gn3/computations/partial_correlations.py
index ba4de9e..1a6868a 100644
--- a/gn3/computations/partial_correlations.py
+++ b/gn3/computations/partial_correlations.py
@@ -9,6 +9,9 @@ from functools import reduce
 from typing import Any, Tuple, Sequence
 from scipy.stats import pearsonr, spearmanr
 
+from gn3.settings import TEXTDIR
+from gn3.data_helpers import parse_csv_line
+
 def control_samples(controls: Sequence[dict], sampleslist: Sequence[str]):
     """
     Fetches data for the control traits.
@@ -192,3 +195,49 @@ def good_dataset_samples_indexes(
     return tuple(sorted(
         samples_from_file.index(good) for good in
         set(samples).intersection(set(samples_from_file))))
+
+def compute_partial_correlations_fast(# pylint: disable=[R0913, R0914]
+        samples, primary_vals, control_vals, database_filename,
+        fetched_correlations, method: str, correlation_type: str) -> Tuple[
+            float, Tuple[float, ...]]:
+    """
+    This is a partial migration of the
+    `web.webqtl.correlation.PartialCorrDBPage.getPartialCorrelationsFast`
+    function in GeneNetwork1.
+    """
+    assert method in ("spearman", "pearson")
+    with open(f"{TEXTDIR}/{database_filename}", "r") as dataset_file:
+        dataset = tuple(dataset_file.readlines())
+
+    good_dataset_samples = good_dataset_samples_indexes(
+        samples, parse_csv_line(dataset[0])[1:])
+
+    def __process_trait_names_and_values__(acc, line):
+        trait_line = parse_csv_line(line)
+        trait_name = trait_line[0]
+        trait_data = trait_line[1:]
+        if trait_name in fetched_correlations.keys():
+            return (
+                acc[0] + (trait_name,),
+                acc[1] + tuple(
+                    trait_data[i] if i in good_dataset_samples else None
+                    for i in range(len(trait_data))))
+        return acc
+
+    processed_trait_names_values: tuple = reduce(
+        __process_trait_names_and_values__, dataset[1:], (tuple(), tuple()))
+    all_target_trait_names: Tuple[str, ...] = processed_trait_names_values[0]
+    all_target_trait_values: Tuple[float, ...] = processed_trait_names_values[1]
+
+    all_correlations = determine_partials(
+        primary_vals, control_vals, all_target_trait_names,
+        all_target_trait_values, method)
+    ## Line 772 to 779 in GN1 are the cause of the weird complexity in the
+    ## return below. Once the surrounding code is successfully migrated and
+    ## reworked, this complexity might go away, by getting rid of the
+    ## `correlation_type` parameter
+    return len(all_correlations), tuple(
+        corr + (
+            (fetched_correlations[corr[0]],) if correlation_type == "literature"
+            else fetched_correlations[corr[0]][0:2])
+        for idx, corr in enumerate(all_correlations))
-- 
cgit v1.2.3