def pool_test_dataset_dir(obs_dir_fp,exp_dir_fp,file_name_delimiter="--",\
file_name_field_order=\
{'file_type':0,"prediction_method":1,"weighting_method":2,"holdout_method":3,\
"distance":4,"organism":5},strict=False, verbose=True,pool_by=['distance']):
"""Retrun pooled control & evaluation results from the given directories
obs_dir_fp -- directory containing PICRUST-predicted genomes. These MUST start with
'predict_traits', and must contain the values specified in file_name_field_order,\
separated by the delimiter given in file_name_delimiter. For example:
predict_traits--exclude_tips_by_distance--0.87--'NC_000913|646311926'
exp_dir_fp -- as obs_dir_fp above, but expectation file names (usually sequenced genomes
with known gene content) must start with exp_biom_traits
file_name_delimiter -- the delimiter that separates metadata stored in the filename
NOTE: technically this isn't the best way of doing things. We may want at some point
to revisit this setup and store metadata about each comparison in a separate file. But
storing in the filename is convenient for our initial analysis.
file_name_field_order -- the order of the required metadata fields in the filename.
Required fields are file_type,method,distance,and organism
pool_by -- if passed, concatenate traits from each trial that is identical in this category. e.g. pool_by 'distance' will pool traits across individual test genomes with the same holdout distance.
The method assumes that for each file type in the observed directory, a paired file
is also found in the exp_dir with similar method, distance, and organism, but a varied
file type (test_tree, test_trait_table)
Process:
1. Search test directory for all gene predictions in the correct format
2. For each, find the corresponding expected trait table in the expectation file
3. Pool by specified pool_by values
4. Return dicts of pooled observation,expectation values
"""
trials = defaultdict(list)
#We'll want a quick unzip fn for converting points to trials
#TODO: separate out into a 'get_paired_data_from_dirs' function
pooled_observations = {}
pooled_expectations = {}
pairs = iter_prediction_expectation_pairs(obs_dir_fp,exp_dir_fp,file_name_field_order,file_name_delimiter,verbose=verbose)
file_number = 0
for obs_table,exp_table,filename in pairs:
#print "analyzing filename:",filename
filename_metadata= get_metadata_from_filename(filename,file_name_field_order,\
file_name_delimiter,verbose=verbose)
#base_tag = '%s\t%s\t' %(filename_metadata['holdout_method'],filename_metadata['prediction_method'])
#tags = [base_tag+'all_results']
if 'file_type' in pool_by:
pool_by.remove('file_type') #we do this manually at the end
combined_tag = ['all']*len(file_name_field_order.keys())
for field in file_name_field_order.keys():
#print combined_tag
#print file_name_field_order
idx = file_name_field_order[field]
#print idx
if field in pool_by:
combined_tag[idx] = filename_metadata[field]
tags=[file_name_delimiter.join(combined_tag)]
if verbose:
print "Pooling by:", pool_by
print "Combined tags:",tags
pooled_observations,pooled_expectations =\
update_pooled_data(obs_table,exp_table,tags,pooled_observations,\
pooled_expectations,str(file_number),verbose=verbose)
file_number += 1
return pooled_observations,pooled_expectations
print "Missing expectation file....skipping!"
continue
base_tag = '%s\t%s\t' %(holdout_method,prediction_method)
tags = [base_tag+'all_results']
combined_tag = base_tag +\
"\t".join([str(field)+"_"+str(filename_components[file_name_field_order[field]]) for field in pool_by])
tags.append(combined_tag)
#if verbose:
# print "Pooling by:", pool_by
# print "Combined tags:",tags
#TODO: abstract out pooling into its own function
non_pooled_fields = [filename_components.get(file_name_field_order[k],None) for k in file_name_field_order.keys() if k not in pool_by]
pooled_observations,pooled_expectations =\
update_pooled_data(obs_table,exp_table,tags,pooled_observations,\
pooled_expectations,str(file_number),verbose=verbose)
#if verbose:
# for tag in pooled_observations.keys():
# print "Merged obs biom:", pooled_observations[tag]
# print "\nMedged *exp* biom:", pooled_expectations[tag]
return run_accuracy_calculations_on_pooled_data(pooled_observations,\
pooled_expectations,roc_success_criteria=roc_success_criteria,verbose=verbose)
def main():
option_parser, opts, args =\
parse_command_line_parameters(**script_info)
pool_by = opts.pool_by.split(',')
def pool_test_dataset_dir(obs_dir_fp,exp_dir_fp,file_name_delimiter="--",\
file_name_field_order=\
{'file_type':0,"prediction_method":1,"weighting_method":2,"holdout_method":3,\
"distance":4,"organism":5},strict=False, verbose=True,pool_by=['distance']):
"""Retrun pooled control & evaluation results from the given directories
obs_dir_fp -- directory containing PICRUST-predicted genomes. These MUST start with
'predict_traits', and must contain the values specified in file_name_field_order,\
separated by the delimiter given in file_name_delimiter. For example:
predict_traits--exclude_tips_by_distance--0.87--'NC_000913|646311926'
exp_dir_fp -- as obs_dir_fp above, but expectation file names (usually sequenced genomes
with known gene content) must start with exp_biom_traits
file_name_delimiter -- the delimiter that separates metadata stored in the filename
NOTE: technically this isn't the best way of doing things. We may want at some point
to revisit this setup and store metadata about each comparison in a separate file. But
storing in the filename is convenient for our initial analysis.
file_name_field_order -- the order of the required metadata fields in the filename.
Required fields are file_type,method,distance,and organism
pool_by -- if passed, concatenate traits from each trial that is identical in this category. e.g. pool_by 'distance' will pool traits across individual test genomes with the same holdout distance.
The method assumes that for each file type in the observed directory, a paired file
is also found in the exp_dir with similar method, distance, and organism, but a varied
file type (test_tree, test_trait_table)
Process:
1. Search test directory for all gene predictions in the correct format
2. For each, find the corresponding expected trait table in the expectation file
3. Pool by specified pool_by values
4. Return dicts of pooled observation,expectation values
"""
#We'll want a quick unzip fn for converting points to trials
#TODO: separate out into a 'get_paired_data_from_dirs' function
pooled_observations = {}
pooled_expectations = {}
pairs = iter_prediction_expectation_pairs(obs_dir_fp,
exp_dir_fp,
file_name_field_order,
file_name_delimiter,
verbose=verbose)
file_number = 0
for obs_table, exp_table, filename in pairs:
#print "analyzing filename:",filename
filename_metadata= get_metadata_from_filename(filename,file_name_field_order,\
file_name_delimiter,verbose=verbose)
#base_tag = '%s\t%s\t' %(filename_metadata['holdout_method'],filename_metadata['prediction_method'])
#tags = [base_tag+'all_results']
if 'file_type' in pool_by:
pool_by.remove('file_type') #we do this manually at the end
combined_tag = ['all'] * len(file_name_field_order.keys())
for field in file_name_field_order.keys():
#print combined_tag
#print file_name_field_order
idx = file_name_field_order[field]
#print idx
if field in pool_by:
combined_tag[idx] = filename_metadata[field]
tags = [file_name_delimiter.join(combined_tag)]
if verbose:
print "Pooling by:", pool_by
print "Combined tags:", tags
pooled_observations,pooled_expectations =\
update_pooled_data(obs_table,exp_table,tags,pooled_observations,\
pooled_expectations,str(file_number),verbose=verbose)
file_number += 1
return pooled_observations, pooled_expectations
if strict:
raise IOError(e)
else:
if verbose:
print "Missing expectation file....skipping!"
continue
base_tag = '%s\t%s\t' %(holdout_method,prediction_method)
tags = [base_tag+'all_results']
combined_tag = base_tag +\
"\t".join([str(field)+"_"+str(filename_components[file_name_field_order[field]]) for field in pool_by])
tags.append(combined_tag)
#TODO: abstract out pooling into its own function
non_pooled_fields = [filename_components.get(file_name_field_order[k],None) for k in file_name_field_order.keys() if k not in pool_by]
pooled_observations,pooled_expectations =\
update_pooled_data(obs_table,exp_table,tags,pooled_observations,\
pooled_expectations,str(file_number),verbose=verbose)
return run_accuracy_calculations_on_pooled_data(pooled_observations,\
pooled_expectations,roc_success_criteria=roc_success_criteria,verbose=verbose)
def main():
option_parser, opts, args =\
parse_command_line_parameters(**script_info)
pool_by = opts.pool_by.split(',')
#create output directory
make_output_dir(opts.output_dir)
#Construct a dict from user specified field order
file_name_field_order = {}
