aboutsummaryrefslogtreecommitdiff
path: root/wqflask/base/data_set.py
diff options
context:
space:
mode:
Diffstat (limited to 'wqflask/base/data_set.py')
-rwxr-xr-xwqflask/base/data_set.py32
1 files changed, 15 insertions, 17 deletions
diff --git a/wqflask/base/data_set.py b/wqflask/base/data_set.py
index 68a2a185..aaea7ec3 100755
--- a/wqflask/base/data_set.py
+++ b/wqflask/base/data_set.py
@@ -280,7 +280,7 @@ class DatasetGroup(object):
"""
def __init__(self, dataset):
"""This sets self.group and self.group_id"""
- print("DATASET NAME2:", dataset.name)
+ #print("DATASET NAME2:", dataset.name)
self.name, self.id = g.db.execute(dataset.query_for_group).fetchone()
if self.name == 'BXD300':
self.name = "BXD"
@@ -366,7 +366,7 @@ class DatasetGroup(object):
break
if tissue_already_exists:
- print("dataset_menu:", dataset_menu[i]['datasets'])
+ #print("dataset_menu:", dataset_menu[i]['datasets'])
dataset_menu[i]['datasets'].append((dataset, dataset_short))
else:
dataset_menu.append(dict(tissue=tissue_name,
@@ -392,18 +392,18 @@ class DatasetGroup(object):
def get_samplelist(self):
key = "samplelist:v2:" + self.name
- print("key is:", key)
- with Bench("Loading cache"):
- result = Redis.get(key)
+ #print("key is:", key)
+ #with Bench("Loading cache"):
+ result = Redis.get(key)
if result:
- print("Sample List Cache hit!!!")
- print("Before unjsonifying {}: {}".format(type(result), result))
+ #print("Sample List Cache hit!!!")
+ #print("Before unjsonifying {}: {}".format(type(result), result))
self.samplelist = json.loads(result)
- print(" type: ", type(self.samplelist))
- print(" self.samplelist: ", self.samplelist)
+ #print(" type: ", type(self.samplelist))
+ #print(" self.samplelist: ", self.samplelist)
else:
- print("Cache not hit")
+ #print("Cache not hit")
from utility.tools import plink_command
PLINK_PATH,PLINK_COMMAND = plink_command()
@@ -417,7 +417,7 @@ class DatasetGroup(object):
self.samplelist = get_group_samplelists.get_samplelist("geno", geno_file_path)
else:
self.samplelist = None
- print("after get_samplelist")
+ #print("after get_samplelist")
Redis.set(key, json.dumps(self.samplelist))
Redis.expire(key, 60*5)
@@ -442,9 +442,9 @@ class DatasetGroup(object):
# reaper barfs on unicode filenames, so here we ensure it's a string
full_filename = str(os.path.join(webqtlConfig.GENODIR, self.name + '.geno'))
if os.path.isfile(full_filename):
- print("Reading file: ", full_filename)
+ #print("Reading file: ", full_filename)
genotype_1.read(full_filename)
- print("File read")
+ #print("File read")
else:
try:
full_filename = str(os.path.join(webqtlConfig.TMPDIR, self.name + '.geno'))
@@ -521,8 +521,6 @@ class DataSet(object):
self.group.get_samplelist()
self.species = species.TheSpecies(self)
- print("TESTING!!!")
-
def get_desc(self):
"""Gets overridden later, at least for Temp...used by trait's get_given_name"""
@@ -700,7 +698,7 @@ class PhenotypeDataSet(DataSet):
def setup(self):
- print("IS A PHENOTYPEDATASET")
+ #print("IS A PHENOTYPEDATASET")
# Fields in the database table
self.search_fields = ['Phenotype.Post_publication_description',
@@ -1032,7 +1030,7 @@ class MrnaAssayDataSet(DataSet):
#print("After get_trait_list query")
trait_data = {}
for trait in results:
- print("Retrieving sample_data for ", trait[0])
+ #print("Retrieving sample_data for ", trait[0])
trait_data[trait[0]] = self.retrieve_sample_data(trait[0])
#print("After retrieve_sample_data")
return trait_data