new_samplefile = swap.get_new_filename(tonights.parameters,'training_false_positives')
print "SWAP: saving false positives..."
N = swap.write_list(sample,new_samplefile,item='false_positive')
print "SWAP: "+str(N)+" lines written to "+new_samplefile
new_samplefile = swap.get_new_filename(tonights.parameters,'training_false_negatives')
print "SWAP: saving false negatives..."
N = swap.write_list(sample,new_samplefile,item='false_negative')
print "SWAP: "+str(N)+" lines written to "+new_samplefile
# Also write out catalogs of subjects, including the ZooID, subject ID,
# how many classifications, and probability:
catalog = swap.get_new_filename(tonights.parameters,'candidate_catalog')
print "SWAP: saving catalog of high probability subjects..."
Nlenses,Nsubjects = swap.write_catalog(sample,catalog,thresholds,kind='test')
print "SWAP: From "+str(Nsubjects)+" subjects classified,"
print "SWAP: "+str(Nlenses)+" candidates (with P > rejection) written to "+catalog
catalog = swap.get_new_filename(tonights.parameters,'sim_catalog')
print "SWAP: saving catalog of high probability subjects..."
Nsims,Nsubjects = swap.write_catalog(sample,catalog,thresholds,kind='sim')
print "SWAP: From "+str(Nsubjects)+" subjects classified,"
print "SWAP: "+str(Nsims)+" sim 'candidates' (with P > rejection) written to "+catalog
catalog = swap.get_new_filename(tonights.parameters,'dud_catalog')
print "SWAP: saving catalog of high probability subjects..."
Nduds,Nsubjects = swap.write_catalog(sample,catalog,thresholds,kind='dud')
print "SWAP: From "+str(Nsubjects)+" subjects classified,"
print "SWAP: "+str(Nduds)+" dud 'candidates' (with P > rejection) written to "+catalog
def make_offline_reports(args):
"""
NAME
make_offline_reports
PURPOSE
Given an offline tuple as well as other bureau tuples etc,
this script produces the reports made at the end of SWAP
COMMENTS
FLAGS
-h Print this message
--out Output directory, otherwise is '.'
--do_offline Do offline analysis?
INPUTS
configfile Plain text file containing SW experiment configuration
bureaufile
collectionfile
OUTPUTS
EXAMPLE
BUGS
AUTHORS
This file is part of the Space Warps project, and is distributed
under the MIT license by the Space Warps Science Team.
http://spacewarps.org/
HISTORY
2014-09-16 started Davis (KIPAC)
"""
# ------------------------------------------------------------------
# Some defaults:
# default settings are for offline using only exact training info
flags = {'do_offline': False,
'output_directory': '.',
'PL0': 0.5, # initial PL guess
'PD0': 0.5, # initial PD guess
'pi': 4e-2, # initial lens probability
'n_min_assessment': 0, # minimum number of assessments before included in analysis
'use_training_info': True,
'exclude_test_info': True,
'exclude_training_info': False,
'N_min': 10, # min number of EM steps required
'N_max': 100, # max number of EM steps
'epsilon_min': 1e-6, # escape condition
}
# this has to be easier to do...
for arg in args:
if arg in flags:
flags[arg] = args[arg]
elif arg == 'config':
configfile = args[arg]
elif arg == 'collection':
collectionfile = args[arg]
elif arg == 'bureau':
bureaufile = args[arg]
else:
print "make_offline_reports: unrecognized flag ",arg
out_dir = flags['output_directory']
# ------------------------------------------------------------------
# Read in run configuration:
tonights = swap.Configuration(configfile)
# TODO: do this correctly
tonights.parameters['finish'] = 'now'
tonights.parameters['start'] = 'now'
tonights.parameters['trunk'] = \
tonights.parameters['survey']+'_'+tonights.parameters['finish']
tonights.parameters['dir'] = out_dir
# How will we make decisions based on probability?
thresholds = {}
thresholds['detection'] = tonights.parameters['detection_threshold']
thresholds['rejection'] = tonights.parameters['rejection_threshold']
t = -1 # for now?!
# ------------------------------------------------------------------
# Read in, or create, a bureau of agents who will represent the
# volunteers:
bureau = swap.read_pickle(bureaufile, 'bureau')
# ------------------------------------------------------------------
# Read in, or create, an object representing the candidate list:
sample = swap.read_pickle(collectionfile, 'collection')
# ------------------------------------------------------------------
# if do_offline, run offline analysis here:
if flags['do_offline']:
PL0 = flags['PL0']
PD0 = flags['PD0']
pi = flags['pi']
n_min_assessment = flags['n_min_assessment']
use_training_info = flags['use_training_info']
exclude_test_info = flags['exclude_test_info']
exclude_training_info = flags['exclude_training_info']
N_min = flags['N_min']
N_max = flags['N_max']
epsilon_min = flags['epsilon_min']
# initialize offline params
bureau_offline = {}
probabilities = {}
online_probabilities = {}
training_IDs = {} # which entries in collection are training
set_aside_subject = {} # which subjects do we set aside? Here we set aside none
set_aside_agent = {} # which agents do we set aside? Here we set aside none
collection = {}
for ID in sample.list():
if ID in set_aside_subject:
continue
else:
collection.update({ID: sample.member[ID]})
for ID in collection.keys():
subject = collection[ID]
n_assessment = len(subject.annotationhistory['ItWas'])
if (n_assessment > n_min_assessment):
if (subject.category == 'training'):
if use_training_info:
truth = {'LENS': 1, 'NOT': 0}[subject.truth]
training_IDs.update({ID: truth})
if exclude_training_info:
# when doing M step, don't use these to update parameters
training_IDs.update({ID: -1})
elif (subject.category == 'test'):
if exclude_test_info:
# when doing M step, don't use these to update parameters
training_IDs.update({ID: -1})
probabilities.update({ID: pi})
online_probabilities.update({ID: subject.mean_probability})
for agent_i in xrange(len(subject.annotationhistory['Name'])):
name = subject.annotationhistory['Name'][agent_i]
if name in set_aside_agent:
continue
xij = subject.annotationhistory['ItWas'][agent_i]
if name not in bureau_offline:
bureau_offline.update({name: {
'PD': PD0, 'PL': PL0,
'PL_in': bureau.member[name].PL,
'PD_in': bureau.member[name].PD,
'Pi': pi,
'Subjects': {ID: xij}}})
else:
bureau_offline[name]['Subjects'].update({ID: xij})
# Run EM Algorithm
bureau_offline, pi, probabilities, information_dict = EM_algorithm(
bureau_offline, pi, probabilities, training_IDs,
N_min=N_min, N_max=N_max, epsilon_min=epsilon_min,
return_information=True)
tup = (bureau_offline, pi, probabilities, information_dict)
offlinefile = out_dir + '/offline.pickle'
swap.write_pickle(tup, offlinefile)
# ------------------------------------------------------------------
# Now replace sample member probabilities with offline probabilities
# Also update bureau with offline results
for ID in sample.list():
# just in case any IDs didn't get into offline somehow?!
if ID not in probabilities.keys():
sample.member.pop(ID)
continue
# This is a bit hackish: update mean_probability,
# median_probability, and do the rejection threshold stuff
subject = sample.member[ID]
subject.mean_probability = probabilities[ID]
subject.median_probability = probabilities[ID]
# ripped from subject.py
if subject.mean_probability < subject.rejection_threshold:
subject.status = 'rejected'
if subject.kind == 'test':
subject.state = 'inactive'
subject.retirement_time = -1#at_time
subject.retirement_age = subject.exposure
elif subject.mean_probability > subject.detection_threshold:
subject.status = 'detected'
if subject.kind == 'test':
# Let's keep the detections live!
# subject.state = 'inactive'
# subject.retirement_time = at_time
# subject.retirement_age = subject.exposure
pass
else:
# Keep the subject alive! This code is only reached if
# we are not being hasty.
subject.status = 'undecided'
if subject.kind == 'test':
subject.state = 'active'
subject.retirement_time = 'not yet'
subject.retirement_age = 0.0
# I don't think this is necessary, but just in case
sample.member[ID] = subject
for kind in ['sim', 'dud', 'test']:
sample.collect_probabilities(kind)
# now save
collectionfile = out_dir + '/collection_offline.pickle'
swap.write_pickle(collection, collectionfile)
# now update bureau
for ID in bureau.list():
# just in case any IDs didn't make it to offline?
if ID not in bureau_offline.keys():
bureau.member.pop(ID)
continue
# update PL, PD, then update_skill
agent = bureau.member[ID]
agent.PL = bureau_offline[ID]['PL']
agent.PD = bureau_offline[ID]['PD']
agent.update_skill()
# I don't think this is necessary, but just in case
bureau.member[ID] = agent
bureau.collect_probabilities()
# now save
bureaufile = out_dir + '/bureau_offline.pickle'
swap.write_pickle(bureau, bureaufile)
# ------------------------------------------------------------------
# now we can pretend we're in SWAP.py
new_retirementfile = swap.get_new_filename(tonights.parameters,'retire_these')
print "make_offline_reports: saving retiree subject Zooniverse IDs..."
N = swap.write_list(sample,new_retirementfile,item='retired_subject')
print "make_offline_reports: "+str(N)+" lines written to "+new_retirementfile
# Also print out lists of detections etc! These are urls of images.
new_samplefile = swap.get_new_filename(tonights.parameters,'candidates')
print "make_offline_reports: saving lens candidates..."
N = swap.write_list(sample,new_samplefile,item='candidate')
print "make_offline_reports: "+str(N)+" lines written to "+new_samplefile
# Now save the training images, for inspection:
new_samplefile = swap.get_new_filename(tonights.parameters,'training_true_positives')
print "make_offline_reports: saving true positives..."
N = swap.write_list(sample,new_samplefile,item='true_positive')
print "make_offline_reports: "+str(N)+" lines written to "+new_samplefile
new_samplefile = swap.get_new_filename(tonights.parameters,'training_false_positives')
print "make_offline_reports: saving false positives..."
N = swap.write_list(sample,new_samplefile,item='false_positive')
print "make_offline_reports: "+str(N)+" lines written to "+new_samplefile
new_samplefile = swap.get_new_filename(tonights.parameters,'training_false_negatives')
print "make_offline_reports: saving false negatives..."
N = swap.write_list(sample,new_samplefile,item='false_negative')
print "make_offline_reports: "+str(N)+" lines written to "+new_samplefile
# Also write out catalogs of subjects, including the ZooID, subject ID,
# how many classifications, and probability:
catalog = swap.get_new_filename(tonights.parameters,'candidate_catalog')
print "make_offline_reports: saving catalog of high probability subjects..."
Nlenses,Nsubjects = swap.write_catalog(sample,catalog,thresholds,kind='test')
print "make_offline_reports: From "+str(Nsubjects)+" subjects classified,"
print "make_offline_reports: "+str(Nlenses)+" candidates (with P > rejection) written to "+catalog
catalog = swap.get_new_filename(tonights.parameters,'sim_catalog')
print "make_offline_reports: saving catalog of high probability subjects..."
Nsims,Nsubjects = swap.write_catalog(sample,catalog,thresholds,kind='sim')
print "make_offline_reports: From "+str(Nsubjects)+" subjects classified,"
print "make_offline_reports: "+str(Nsims)+" sim 'candidates' (with P > rejection) written to "+catalog
catalog = swap.get_new_filename(tonights.parameters,'dud_catalog')
print "make_offline_reports: saving catalog of high probability subjects..."
Nduds,Nsubjects = swap.write_catalog(sample,catalog,thresholds,kind='dud')
print "make_offline_reports: From "+str(Nsubjects)+" subjects classified,"
print "make_offline_reports: "+str(Nduds)+" dud 'candidates' (with P > rejection) written to "+catalog
# ------------------------------------------------------------------
# Make plots! Can't plot everything - uniformly sample 200 of each
# thing (agent or subject).
# Agent histories:
fig1 = bureau.start_history_plot()
pngfile = swap.get_new_filename(tonights.parameters,'histories')
Nc = np.min([200,bureau.size()])
print "make_offline_reports: plotting "+str(Nc)+" agent histories in "+pngfile
for Name in bureau.shortlist(Nc):
bureau.member[Name].plot_history(fig1)
bureau.finish_history_plot(fig1,t,pngfile)
tonights.parameters['historiesplot'] = pngfile
# Agent probabilities:
pngfile = swap.get_new_filename(tonights.parameters,'probabilities')
print "make_offline_reports: plotting "+str(Nc)+" agent probabilities in "+pngfile
bureau.plot_probabilities(Nc,t,pngfile)
tonights.parameters['probabilitiesplot'] = pngfile
# Subject trajectories:
fig3 = sample.start_trajectory_plot()
pngfile = swap.get_new_filename(tonights.parameters,'trajectories')
# Random 500 for display purposes:
Ns = np.min([500,sample.size()])
print "make_offline_reports: plotting "+str(Ns)+" subject trajectories in "+pngfile
for ID in sample.shortlist(Ns):
sample.member[ID].plot_trajectory(fig3)
# To plot only false negatives, or only true positives:
# for ID in sample.shortlist(Ns,kind='sim',status='rejected'):
# sample.member[ID].plot_trajectory(fig3)
# for ID in sample.shortlist(Ns,kind='sim',status='detected'):
# sample.member[ID].plot_trajectory(fig3)
sample.finish_trajectory_plot(fig3,pngfile,t=t)
tonights.parameters['trajectoriesplot'] = pngfile
# Candidates! Plot all undecideds or detections:
fig4 = sample.start_trajectory_plot(final=True)
pngfile = swap.get_new_filename(tonights.parameters,'sample')
# BigN = 100000 # Would get them all...
BigN = 500 # Can't see them all!
candidates = []
candidates += sample.shortlist(BigN,kind='test',status='detected')
candidates += sample.shortlist(BigN,kind='test',status='undecided')
sims = []
sims += sample.shortlist(BigN,kind='sim',status='detected')
sims += sample.shortlist(BigN,kind='sim',status='undecided')
duds = []
duds += sample.shortlist(BigN,kind='dud',status='detected')
duds += sample.shortlist(BigN,kind='dud',status='undecided')
print "make_offline_reports: plotting "+str(len(sims))+" sims in "+pngfile
for ID in sims:
sample.member[ID].plot_trajectory(fig4)
print "make_offline_reports: plotting "+str(len(duds))+" duds in "+pngfile
for ID in duds:
sample.member[ID].plot_trajectory(fig4)
print "make_offline_reports: plotting "+str(len(candidates))+" candidates in "+pngfile
for ID in candidates:
sample.member[ID].plot_trajectory(fig4)
# They will all show up in the histogram though:
sample.finish_trajectory_plot(fig4,pngfile,final=True)
tonights.parameters['candidatesplot'] = pngfile
# ------------------------------------------------------------------
# Finally, write a PDF report:
swap.write_report(tonights.parameters,bureau,sample)
print "SWAP: " + str(N) + " lines written to " + new_samplefile
new_samplefile = swap.get_new_filename(tonights.parameters,
'training_false_negatives')
print "SWAP: saving false negatives..."
N = swap.write_list(sample, new_samplefile, item='false_negative')
print "SWAP: " + str(N) + " lines written to " + new_samplefile
# Also write out catalogs of subjects, including the ZooID, subject ID,
# how many classifications, and probability:
catalog = swap.get_new_filename(tonights.parameters,
'candidate_catalog')
print "SWAP: saving catalog of high probability subjects..."
Nlenses, Nsubjects = swap.write_catalog(sample,
catalog,
thresholds,
kind='test')
print "SWAP: From " + str(Nsubjects) + " subjects classified,"
print "SWAP: " + str(
Nlenses) + " candidates (with P > rejection) written to " + catalog
catalog = swap.get_new_filename(tonights.parameters, 'sim_catalog')
print "SWAP: saving catalog of high probability subjects..."
Nsims, Nsubjects = swap.write_catalog(sample,
catalog,
thresholds,
kind='sim')
print "SWAP: From " + str(Nsubjects) + " subjects classified,"
print "SWAP: " + str(
Nsims
) + " sim 'candidates' (with P > rejection) written to " + catalog
N = swap.write_list(sample,new_samplefile,item='false_positive')
print "SWAP: "+str(N)+" lines written to "+new_samplefile
new_samplefile = swap.get_new_filename(tonights.parameters,\
'training_false_negatives')
print "SWAP: saving false negatives..."
N = swap.write_list(sample,new_samplefile,item='false_negative')
print "SWAP: "+str(N)+" lines written to "+new_samplefile
# -------------------------------------------------------------------
##### THESE ARE CATALOGS THAT SPACEWARPS WAS INTERESTED IN #####
catalog = swap.get_new_filename(tonights.parameters,'candidate_catalog')
print "SWAP: saving catalog of high probability subjects..."
N, Ntot = swap.write_catalog(sample,catalog,thresholds, kind='test')
print "SWAP: From "+str(Ntot)+" subjects classified,"
print "SWAP: "+str(N)+" candidates (with P > rejection) written to "\
+catalog
catalog = swap.get_new_filename(tonights.parameters,'sim_catalog')
print "SWAP: saving catalog of high probability subjects..."
N, Ntot = swap.write_catalog(sample,catalog,thresholds, kind='sim')
print "SWAP: From "+str(Ntot)+" subjects classified,"
print "SWAP: "+str(N)+" sim 'candidates' (with P > rejection) "\
"written to "+catalog
catalog = swap.get_new_filename(tonights.parameters,'dud_catalog')
print "SWAP: saving catalog of high probability subjects..."
N,Ntot = swap.write_catalog(sample,catalog,thresholds,kind='dud')
print "SWAP: From "+str(Ntot)+" subjects classified,"
def MachineClassifier(options, args):
try: config = options.configfile
except: pdb.set_trace()
tonights = swap.Configuration(config)
#"""
# Read the pickled random state file
random_file = open(tonights.parameters['random_file'],"r");
random_state = cPickle.load(random_file);
random_file.close();
np.random.set_state(random_state);
#"""
# Get the machine threshold (make retirement decisions)
threshold = tonights.parameters['machine_threshold']
# Get list of evaluation metrics and criteria
eval_metrics = tonights.parameters['evaluation_metrics']
survey = tonights.parameters['survey']
subdir = 'sup_run4'
#----------------------------------------------------------------------
# read in the metadata for all subjects (Test or Training sample?)
subjects = swap.read_pickle(tonights.parameters['metadatafile'], 'metadata')
#----------------------------------------------------------------------
# read in the SWAP collection
sample = swap.read_pickle(tonights.parameters['samplefile'],'collection')
#----------------------------------------------------------------------
# read in or create the ML collection
MLsample = swap.read_pickle(tonights.parameters['MLsamplefile'],
'MLcollection')
# read in or create the ML bureau for machine agents (history)
MLbureau = swap.read_pickle(tonights.parameters['MLbureaufile'], 'MLbureau')
#-----------------------------------------------------------------------
# DETERMINE IF THERE IS A TRAINING SAMPLE TO WORK WITH
#-----------------------------------------------------------------------
# TO DO: training sample should only select those which are NOT part of
# validation sample (Nair catalog objects) 2/22/16
# IDENTIFY TRAINING SAMPLE
train_sample = subjects[subjects['MLsample']=='train']
train_meta, train_features = ml.extract_training(train_sample)
train_labels = np.array([1 if p > 0.3 else 0 \
for p in train_meta['SWAP_prob']])
# IDENTIFY VALIDATION SAMPLE (FINAL)
valid_sample = subjects[subjects['MLsample']=='valid']
valid_meta, valid_features = ml.extract_training(valid_sample)
valid_labels = valid_meta['Expert_label'].filled()
#if len(train_sample) >= 100:
# TO DO: LOOP THROUGH DIFFERENT MACHINES? HOW MANY MACHINES?
for metric in eval_metrics:
# REGISTER Machine Classifier
# Construct machine name --> Machine+Metric? For now: KNC
machine = 'KNC'
Name = machine+'_'+metric
# register an Agent for this Machine
try:
test = MLbureau.member[Name]
except:
MLbureau.member[Name] = swap.Agent_ML(Name, metric)
#---------------------------------------------------------------
# TRAIN THE MACHINE; EVALUATE ON VALIDATION SAMPLE
#---------------------------------------------------------------
# Now we run the machine -- need cross validation on whatever size
# training sample we have ..
# For now this will be fixed until we build in other machine options
params = {'n_neighbors':np.arange(1, 2*(len(train_sample)-1) / 3, 2),
'weights':('uniform','distance')}
# Create the model
general_model = GridSearchCV(estimator=KNC(), param_grid=params,
error_score=0, scoring=metric)
# Train the model -- k-fold cross validation is embedded
trained_model = general_model.fit(train_features, train_labels)
# Test "accuracy" (metric of choice) on validation sample
score = trained_model.score(valid_features, valid_labels)
MLbureau.member[Name].record_training(\
model_described_by=trained_model.best_estimator_,
with_params=trained_model.best_params_,
trained_on=len(train_features),
at_time=TIME,
with_train_acc=traineed_model.best_score_,
and_valid_acc=trained_model.score(valid_features,
valid_labels))
# Store the trained machine
MLbureau.member[Name].model = trained_model
# Compute / store confusion matrix as a function of threshold
# produced by this machine on the Expert Validation sample
fps, tps, thresh = mtrx._binary_clf_curve(valid_labels,
trained_model.predict_proba(valid_features)[:,1])
metric_list = mtrx.compute_binary_metrics(fps, tps)
ACC, TPR, FPR, FNR, TNR, PPV, FDR, FOR, NPV = metric_list
MLbureau.member[Name].record_evaluation(accuracy=ACC,
completeness_s=TPR,
contamination_s=FDR,
completeness_f=TNR,
contamination_f=NPV)
pdb.set_trace()
# 3. compare the metric of choice with the evaluation criterion to
# see if this machine has sufficiently learned?
# ... what if my criterion is simply "Maximize Accuracy"?
# ... or minimize feature contamination? these require that we
# compare tonight's machine with the previous night's machine
# But if my criterion is simply "have feature contam less than 20%"
# then it's easy....
# IF TRAINED MACHINE PREDICTS WELL ON VALIDATION ....
if MLbureau.member[Name].evaluate():
#---------------------------------------------------------------
# APPLY MACHINE TO TEST SAMPLE
#---------------------------------------------------------------
# This requires that my runKNC function returns the Machine Object
shitski=5
#---------------------------------------------------------------
# PROCESS PREDICTIONS/PROBS
#---------------------------------------------------------------
for s,p,l in zip(test_meta, probas, predictions):
ID = str(s['id'])
descriptions = Nair_or_Not(s)
category, kind, flavor, truth = descriptions
# LOAD EACH TEST SUBJECT INTO MACHINE COLLECTION
# -------------------------------------------------------------
try:
test = MLsample.member[ID]
except: MLsample.member[ID] = swap.Subject_ML(ID,
str(s['name']), category, kind,
truth,threshold,s['external_ref'])
tstring = datetime.now().strftime('%Y-%m-%d_%H:%M:%S')
MLsample.member[ID].was_described(by='knn', as_being=1,
withp=p, at_time=tstring)
# NOTE: if subject is Nair (training) it doesn't get flagged as
# inactive but it can be flagged as detected/rejected
# IF MACHINE P >= THRESHOLD, INSERT INTO SWAP COLLECTION
# -------------------------------------------------------------
thresholds = {'detection':0.,'rejection':0.}
if (p >= threshold) or (1-p >= threshold):
print "BOOM! WE'VE GOT A MACHINE-CLASSIFIED SUBJECT:"
print "Probability:",p
# Initialize the subject in SWAP Collection
sample.member[ID] = swap.Subject(ID, str(s['name']),
category, kind,flavor,truth,
thresholds, s['external_ref'],0.)
sample.member[ID].retiredby = 'machine'
# Flag subject as 'INACTIVE' / 'DETECTED' / 'REJECTED'
# ----------------------------------------------------------
if p >= threshold:
sample.member[str(s['id'])].state = 'inactive'
elif 1-p >= threshold:
sample.member[str(s['id'])].status = 'rejected'
#---------------------------------------------------------------
# SAVE MACHINE METADATA?
#---------------------------------------------------------------
print "Size of SWAP sample:", sample.size()
print "Size of ML sample:", MLsample.size()
if tonights.parameters['report']:
# Output list of subjects to retire, based on this batch of
# classifications. Note that what is needed here is the ZooID,
# not the subject ID:
new_retirementfile = swap.get_new_filename(tonights.parameters,\
'retire_these', source='ML')
print "SWAP: saving Machine-retired subject Zooniverse IDs..."
N = swap.write_list(MLsample,new_retirementfile,
item='retired_subject', source='ML')
print "SWAP: "+str(N)+" lines written to "+new_retirementfile
# write catalogs of smooth/not over MLthreshold
# -------------------------------------------------------------
catalog = swap.get_new_filename(tonights.parameters,
'retired_catalog', source='ML')
print "SWAP: saving catalog of Machine-retired subjects..."
Nretired, Nsubjects = swap.write_catalog(MLsample,bureau,
catalog, threshold,
kind='rejected', source='ML')
print "SWAP: From "+str(Nsubjects)+" subjects classified,"
print "SWAP: "+str(Nretired)+" retired (with P < rejection) "\
"written to "+catalog
catalog = swap.get_new_filename(tonights.parameters,
'detected_catalog', source='ML')
print "SWAP: saving catalog of Machine detected subjects..."
Ndetected, Nsubjects = swap.write_catalog(MLsample, bureau,
catalog, threshold,
kind='detected', source='ML')
print "SWAP: From "+str(Nsubjects)+" subjects classified,"
print "SWAP: %i detected (with P > MLthreshold) "\
"written to %s"%(Ndetected, catalog)
# If is hasn't been done already, save the current directory
# ---------------------------------------------------------------------
tonights.parameters['dir'] = os.getcwd()+'/'+tonights.parameters['trunk']
if not os.path.exists(tonights.parameters['dir']):
os.makedirs(tonights.parameters['dir'])
# Repickle all the shits
# -----------------------------------------------------------------------
if tonights.parameters['repickle']:
new_samplefile = swap.get_new_filename(tonights.parameters,'collection')
print "SWAP: saving SWAP subjects to "+new_samplefile
swap.write_pickle(sample,new_samplefile)
tonights.parameters['samplefile'] = new_samplefile
new_samplefile=swap.get_new_filename(tonights.parameters,'MLcollection')
print "SWAP: saving test sample subjects to "+new_samplefile
swap.write_pickle(MLsample,new_samplefile)
tonights.parameters['MLsamplefile'] = new_samplefile
metadatafile = swap.get_new_filename(tonights.parameters,'metadata')
print "SWAP: saving metadata to "+metadatafile
swap.write_pickle(subjects,metadatafile)
tonights.parameters['metadatafile'] = metadatafile
# Update the time increment for SWAP's next run
# -----------------------------------------------------------------------
t2 = datetime.datetime.strptime(tonights.parameters['start'],
'%Y-%m-%d_%H:%M:%S') + \
datetime.timedelta(days=tonights.parameters['increment'])
tstop = datetime.datetime.strptime(tonights.parameters['end'],
'%Y-%m-%d_%H:%M:%S')
if t2 == tstop:
plots = True
else:
tonights.parameters['start'] = t2.strftime('%Y-%m-%d_%H:%M:%S')
# Update configfile to reflect Machine additions
# -----------------------------------------------------------------------
configfile = 'update.config'
random_file = open(tonights.parameters['random_file'],"w");
random_state = np.random.get_state();
cPickle.dump(random_state,random_file);
random_file.close();
swap.write_config(configfile, tonights.parameters)
pdb.set_trace()
N = swap.write_list(sample,new_samplefile,item='false_positive')
print "SWAP: "+str(N)+" lines written to "+new_samplefile
new_samplefile = swap.get_new_filename(tonights.parameters,\
'training_false_negatives')
print "SWAP: saving false negatives..."
N = swap.write_list(sample,new_samplefile,item='false_negative')
print "SWAP: "+str(N)+" lines written to "+new_samplefile
# -------------------------------------------------------------------
##### THESE ARE CATALOGS THAT SPACEWARPS WAS INTERESTED IN #####
catalog = swap.get_new_filename(tonights.parameters,'candidate_catalog')
print "SWAP: saving catalog of high probability subjects..."
N, Ntot = swap.write_catalog(sample,catalog,thresholds, kind='test')
print "SWAP: From "+str(Ntot)+" subjects classified,"
print "SWAP: "+str(N)+" candidates (with P > rejection) written to "\
+catalog
catalog = swap.get_new_filename(tonights.parameters,'sim_catalog')
print "SWAP: saving catalog of high probability subjects..."
N, Ntot = swap.write_catalog(sample,catalog,thresholds, kind='sim')
print "SWAP: From "+str(Ntot)+" subjects classified,"
print "SWAP: "+str(N)+" sim 'candidates' (with P > rejection) "\
"written to "+catalog
catalog = swap.get_new_filename(tonights.parameters,'dud_catalog')
print "SWAP: saving catalog of high probability subjects..."
N,Ntot = swap.write_catalog(sample,catalog,thresholds,kind='dud')
print "SWAP: From "+str(Ntot)+" subjects classified,"
