aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--wqflask/wqflask/marker_regression/run_mapping.py64
1 files changed, 15 insertions, 49 deletions
diff --git a/wqflask/wqflask/marker_regression/run_mapping.py b/wqflask/wqflask/marker_regression/run_mapping.py
index 891fcc66..a1f87b61 100644
--- a/wqflask/wqflask/marker_regression/run_mapping.py
+++ b/wqflask/wqflask/marker_regression/run_mapping.py
@@ -71,56 +71,22 @@ class RunMapping(object):
all_samples_ordered = self.dataset.group.all_samples_ordered()
self.vals = []
- if 'samples' in start_vars:
- self.samples = start_vars['samples'].split(",")
- if (len(genofile_samplelist) != 0):
- for sample in genofile_samplelist:
- if sample in self.samples:
- value = start_vars.get('value:' + sample)
- if value:
- self.vals.append(value)
- else:
- self.vals.append("x")
- else:
- for sample in self.samples:
- value = start_vars.get('value:' + sample)
- if value:
- self.vals.append(value)
+ self.samples = []
+ self.sample_vals = start_vars['sample_vals']
+ sample_val_dict = json.loads(self.sample_vals)
+ samples = sample_val_dict.keys()
+ if (len(genofile_samplelist) != 0):
+ for sample in genofile_samplelist:
+ self.samples.append(sample)
+ if sample in samples:
+ self.vals.append(sample_val_dict[sample])
+ else:
+ self.vals.append("x")
else:
- self.samples = []
- if (len(genofile_samplelist) != 0):
- for sample in genofile_samplelist:
- if sample in self.dataset.group.samplelist:
- in_trait_data = False
- for item in self.this_trait.data:
- if self.this_trait.data[item].name == sample:
- value = start_vars['value:' + self.this_trait.data[item].name]
- self.samples.append(self.this_trait.data[item].name)
- self.vals.append(value)
- in_trait_data = True
- break
- if not in_trait_data:
- value = start_vars.get('value:' + sample)
- if value:
- self.samples.append(sample)
- self.vals.append(value)
- else:
- self.vals.append("x")
- else:
- for sample in self.dataset.group.samplelist: # sample is actually the name of an individual
- in_trait_data = False
- for item in self.this_trait.data:
- if self.this_trait.data[item].name == sample:
- value = start_vars['value:' + self.this_trait.data[item].name]
- self.samples.append(self.this_trait.data[item].name)
- self.vals.append(value)
- in_trait_data = True
- break
- if not in_trait_data:
- value = start_vars.get('value:' + sample)
- if value:
- self.samples.append(sample)
- self.vals.append(value)
+ for sample in self.dataset.group.samplelist:
+ if sample in samples:
+ self.vals.append(sample_val_dict[sample])
+ self.samples.append(sample)
if 'n_samples' in start_vars:
self.n_samples = start_vars['n_samples']