From 982a2323c0b10011bec622d636fca370d8db5480 Mon Sep 17 00:00:00 2001 From: Frederick Muriuki Muriithi Date: Wed, 15 Nov 2023 12:19:36 +0300 Subject: Bug: Insert ProbeSets before average data and cross-refs Insert the ProbeSets first before inserting the average data and cross-referencing it. --- scripts/insert_data.py | 120 +++++++++++++++++++++++++++++++------------------ 1 file changed, 77 insertions(+), 43 deletions(-) (limited to 'scripts') diff --git a/scripts/insert_data.py b/scripts/insert_data.py index b3e9eea..56d880b 100644 --- a/scripts/insert_data.py +++ b/scripts/insert_data.py @@ -3,8 +3,8 @@ import sys import string import random import argparse -from typing import Tuple from functools import reduce +from typing import Tuple, Iterator import MySQLdb as mdb from redis import Redis @@ -59,22 +59,22 @@ def strains_info( return {strain["Name"]: strain for strain in cursor.fetchall()} def read_datavalues(filepath, headings, strain_info): - "Read data values from file" - for row in ( - dict(zip(headings, line)) - for line in read_file_contents(filepath)): - for sname in headings[1:]: - yield { - "ProbeSetId": int(row["ProbeSetID"]), - "StrainId": strain_info[sname]["Id"], - "DataValue": float(row[sname]) - } + from quality_control.debug import __pk__ + return { + str(row["ProbeSetID"]): tuple({ + "ProbeSetName": str(row["ProbeSetID"]), + "StrainId": strain_info[sname]["Id"], + "DataValue": float(row[sname]) + } for sname in headings[1:]) + for row in + (dict(zip(headings, line)) for line in read_file_contents(filepath)) + } def read_probesets(filepath, headings): """Read the ProbeSet names.""" for row in (dict(zip(headings, line)) for line in read_file_contents(filepath)): - yield {"Name": int(row["ProbeSetID"])} + yield {"Name": str(row["ProbeSetID"])} def last_data_id(dbconn: mdb.Connection) -> int: "Get the last id from the database" @@ -138,21 +138,50 @@ def __format_query__(query, params): "INSERT INTO ", "INSERT INTO\n\t") return f"{insert_str}\nVALUES\n\t{values_str};" -def __xref_params__( - dbconn: mdb.Connection, means: tuple[dict, ...]) -> tuple[dict, ...]: - """Process params for cross-reference table.""" - xref_names = tuple({mean["ProbeSetId"] for mean in means}) +def insert_probesets(filepath: str, + dbconn: mdb.Connection, + platform_id: int, + headings: tuple[str, ...], + session_rand_str: str) -> tuple[str, ...]: + probeset_query = ( + "INSERT INTO ProbeSet(ChipId, Name) " + "VALUES (%(ChipId)s, %(Name)s) ") + the_probesets = ({ + **row, + "Name": f"{row['Name']}{session_rand_str}", + "ChipId": platform_id + } for row in read_probesets(filepath, headings)) + probeset_names = tuple() with dbconn.cursor(cursorclass=DictCursor) as cursor: - params_str = ", ".join(["%s"] * len(xref_names)) + while True: + probeset_params = tuple(take(the_probesets, 10000)) + if not bool(probeset_params): + break + print(__format_query__(probeset_query, probeset_params)) + print() + cursor.executemany(probeset_query, probeset_params) + probeset_names = probeset_names + tuple( + name[0:name.index("::RAND_")] for name in ( + row["Name"] for row in probeset_params)) + + return probeset_names + +def probeset_ids(dbconn: mdb.Connection, + chip_id: int, + probeset_names: tuple[str, ...]) -> Iterator[tuple[str, int]]: + """Fetch the IDs of the probesets with the given names.""" + with dbconn.cursor() as cursor: + params_str = ", ".join(["%s"] * len(probeset_names)) cursor.execute( - f"SELECT Name, Id FROM ProbeSet WHERE Name IN ({params_str})", - xref_names) - ids = {row["Name"]: row["Id"] for row in cursor.fetchall()} - return tuple({ - **mean, - "ProbeSetName": mean["ProbeSetId"], - "ProbeSetId": ids[str(mean["ProbeSetId"])] - } for mean in means) + "SELECT Name, Id FROM ProbeSet " + "WHERE ChipId=%s " + f"AND Name IN ({params_str})", + (chip_id,) + probeset_names) + while True: + row = cursor.fetchone() + if not bool(row): + break + yield row def insert_means(# pylint: disable=[too-many-locals, too-many-arguments] filepath: str, speciesid: int, platform_id: int, datasetid: int, @@ -161,40 +190,45 @@ def insert_means(# pylint: disable=[too-many-locals, too-many-arguments] headings = read_file_headings(filepath) strains = strains_info(dbconn, headings[1:], speciesid) check_strains(headings[1:], strains) - probeset_query = ( - "INSERT INTO ProbeSet(ChipId, Name) " - "VALUES (%(ChipId)s, %(Name)s) ") means_query = ( "INSERT INTO ProbeSetData " "VALUES(%(ProbeSetDataId)s, %(StrainId)s, %(DataValue)s)") xref_query = ( "INSERT INTO ProbeSetXRef(ProbeSetFreezeId, ProbeSetId, DataId) " "VALUES(%(ProbeSetFreezeId)s, %(ProbeSetId)s, %(ProbeSetDataId)s)") + + # A random string to avoid over-write chances. + # This is needed because the `ProbeSet` table is defined with + # UNIQUE KEY `ProbeSetId` (`ChipId`,`Name`) + # which means that we cannot have 2 (or more) ProbeSets which share both + # the name and chip_id (platform) at the same time. + rand_str = f"::RAND_{random_string()}" + pset_ids = { + name: pset_id + for name, pset_id in probeset_ids( + dbconn, + platform_id, + insert_probesets( + filepath, dbconn, platform_id, headings, rand_str)) + } the_means = ({ - "ProbeSetFreezeId": datasetid, "ProbeSetDataId": data_id, - "ChipId": platform_id, **mean - } for data_id, mean in enumerate( - read_datavalues(filepath, headings, strains), - start=(last_data_id(dbconn)+1))) - the_probesets = ({ - **row, - "Name": f"{row['Name']}::RAND_{random_string()}", - "ChipId": platform_id - } for row in read_probesets(filepath, headings)) + **mean, "ProbeSetFreezeId": datasetid, "ProbeSetDataId": data_id, + "ChipId": platform_id, "ProbeSetId": pset_ids[mean["ProbeSetName"]] + } for data_id, mean in enumerate(( + item for sublist in + read_datavalues(filepath, headings, strains).values() + for item in sublist), + start=(last_data_id(dbconn)+1))) with dbconn.cursor(cursorclass=DictCursor) as cursor: while True: means = tuple(take(the_means, 10000)) - probeset_params = tuple(take(the_probesets, 10000)) if not bool(means): break - print(__format_query__(probeset_query, probeset_params)) - print() print(__format_query__(means_query, means)) print() print(__format_query__(xref_query, means)) - cursor.executemany(probeset_query, probeset_params) cursor.executemany(means_query, means) - cursor.executemany(xref_query, __xref_params__(dbconn, means)) + cursor.executemany(xref_query, means) return 0 def insert_se(# pylint: disable = [too-many-arguments] -- cgit v1.2.3