diff options
author | Zachary Sloan | 2013-04-09 23:18:05 +0000 |
---|---|---|
committer | Zachary Sloan | 2013-04-09 23:18:05 +0000 |
commit | d2e64633a40ecab55c9cedd46b0a8f9e93761aad (patch) | |
tree | 78c4a182bbede6d431ec9fa11c4c682e7c5508e6 /wqflask/maintenance/quick_search_table.py | |
parent | c7c306c69254ca49ddeccc495a8a096fcf03974d (diff) | |
download | genenetwork2-d2e64633a40ecab55c9cedd46b0a8f9e93761aad.tar.gz |
Have quick search working for phenotypes
Diffstat (limited to 'wqflask/maintenance/quick_search_table.py')
-rw-r--r-- | wqflask/maintenance/quick_search_table.py | 133 |
1 files changed, 123 insertions, 10 deletions
diff --git a/wqflask/maintenance/quick_search_table.py b/wqflask/maintenance/quick_search_table.py index 48697e58..a6ca6265 100644 --- a/wqflask/maintenance/quick_search_table.py +++ b/wqflask/maintenance/quick_search_table.py @@ -88,7 +88,7 @@ class PublishXRef(Base): "PublishXRef.PhenotypeId = Phenotype.Id and " "PublishXRef.PublicationId = Publication.Id ").params(publishxref_id=publishxref_id, inbredset_id=inbredset_id).all() - + unique = set() for item in results[0]: #print("locals:", locals()) @@ -110,11 +110,11 @@ class PublishXRef(Base): print("\n-- UDE \n") # Can't get it into utf-8, we won't use it continue - + unique.add(token) print("\nUnique terms are: {}\n".format(unique)) return " ".join(unique) - + @staticmethod def get_result_fields(publishxref_id, inbredset_id): results = Session.query( @@ -153,9 +153,9 @@ class PublishXRef(Base): # inbredset_id=inbredset_id).all() for result in results: print("****", result) - + assert len(set(result for result in results)) == 1, "Different results or no results" - + print("results are:", results) result = results[0] result = row2dict(result) @@ -170,10 +170,123 @@ class PublishXRef(Base): result[key] = value.decode('utf-8', errors='ignore') json_results = json.dumps(result, sort_keys=True) - return json_results - + return json_results + +class GenoXRef(Base): + __tablename__ = 'ProbeSetXRef' + GenoFreezeId = sa.Column(sa.Integer, primary_key=True) + GenoId = sa.Column(sa.Integer, primary_key=True) + DataId = sa.Column(sa.Integer) + cM = sa.Column(sa.Float) + Used_for_mapping = sa.Column(sa.Text) + + @classmethod + def run(cls): + conn = Engine.connect() + counter = 0 + for item in page_query(Session.query(cls)): #all() + values = {} + values['table_name'] = cls.__tablename__ + values['the_key'] = json.dumps([item.GenoId, item.GenoFreezeId]) + values['terms'] = cls.get_unique_terms(item.GenoId) + print("terms is:", values['terms']) + values['result_fields'] = cls.get_result_fields(item.GenoId, item.GenoFreezeId) + ins = QuickSearch.insert().values(**values) + conn.execute(ins) + counter += 1 + print("Done:", counter) + @staticmethod + def get_unique_terms(geno_id): + results = Session.query( + "name", + "marker_name" + ).from_statement( + "SELECT Geno.Name as name, " + "Geno.Marker_Name as marker_name " + "FROM Geno " + "WHERE Geno.Id = :geno_id ").params(geno_id=geno_id).all() + + unique = set() + for item in results[0]: + #print("locals:", locals()) + if not item: + continue + for token in item.split(): + if len(token) > 2: + try: + # This hopefully ensures that the token is utf-8 + token = token.encode('utf-8') + print(" ->", token) + except UnicodeDecodeError: + print("\n-- UDE \n") + # Can't get it into utf-8, we won't use it + continue + + unique.add(token) + print("\nUnique terms are: {}\n".format(unique)) + return " ".join(unique) + + + @staticmethod + def get_result_fields(geno_id, dataset_id): + results = Session.query( + "name", + "species", + "group_name", + "dataset", + "dataset_name", + "symbol", + "description", + "chr", "mb", + "lrs", + "genbank_id", + "gene_id", + "chip_id", + "chip_name" + ).from_statement( + "SELECT Geno.Name as name, " + "Geno.Marker_Name as marker_name, " + "InbredSet.Name as group_name, " + "Species.Name as species, " + "GenoFreeze.Name as dataset, " + "GenoFreeze.FullName as dataset_name, " + "Geno.Chr as chr, " + "Geno.Mb as mb, " + "Geno.Source as source " + "FROM Geno, " + "GenoXRef, " + "GenoFreeze, " + "InbredSet, " + "Species " + "WHERE Geno.Id = :geno_id and " + "GenoXRef.GenoId = Geno.Id and " + "GenoFreeze.Id = :dataset_id and " + "GenoXRef.GenoFreezeId = GenoFreeze.Id and " + "InbredSet.Id = GenoFreeze.InbredSetId and " + "InbredSet.SpeciesId = Species.Id ").params(geno_id=geno_id, + dataset_id=dataset_id).all() + for result in results: + print(result) + assert len(set(result for result in results)) == 1, "Different results" + + print("results are:", results) + result = results[0] + result = row2dict(result) + try: + json_results = json.dumps(result, sort_keys=True) + except UnicodeDecodeError: + print("\n\nTrying to massage unicode\n\n") + for key, value in result.iteritems(): + print("\tkey is:", key) + print("\tvalue is:", value) + if isinstance(value, basestring): + result[key] = value.decode('utf-8', errors='ignore') + json_results = json.dumps(result, sort_keys=True) + + return json_results + class ProbeSetXRef(Base): __tablename__ = 'ProbeSetXRef' @@ -255,7 +368,7 @@ class ProbeSetXRef(Base): results = Session.query( "name", "species", - "group", + "group_name", "dataset", "dataset_name", "symbol", @@ -269,7 +382,7 @@ class ProbeSetXRef(Base): ).from_statement( "SELECT ProbeSet.Name as name, " "Species.Name as species, " - "InbredSet.Name as group, " + "InbredSet.Name as group_name, " "ProbeSetFreeze.Name as dataset, " "ProbeSetFreeze.FullName as dataset_name, " "ProbeSet.Symbol as symbol, " @@ -350,8 +463,8 @@ def page_query(q): def main(): - PublishXRef.run() ProbeSetXRef.run() + PublishXRef.run() if __name__ == "__main__": main()
\ No newline at end of file |