def learn_data_grid_general(hybdatadict, SDparams,prb,detectioncrit,supervised_params,addtokwik):
'''
If addtokwik == True, then the clusterings are also stored in the .kwik file
calls learn_data() for various values of the
grids and also the function compute_errors()
Writes output as clusterings labelled by Hash(svmparams) of the grid in
Hash(hybdatadict)_Hash(sdparams)_Hash(detectioncrit)_Hash(supervised_params).kwik
using write_kwik(hybdatadict,sdparams,detectioncrit,svmparams,confusion_test,confusion_train)
the new .kwik format can store multiple clusterings.
supervised_params consists of the following quantities:
supervised_params = {'numfirstspikes': 200000,'kernel': 'rbf','grid_C': [1,100000,0.00001], 'grid_weights': listofweights
,gammagrid : [1e-5, 0.001, 0.1, 1, 10, 1000, 100000], cross_param : 2,
PCAS : 3, subvector: None}
'''
#----------------------------------------------------------
argPLDG = [hybdatadict, SDparams,prb,detectioncrit,supervised_params]
if ju.is_cached(pre_learn_data_grid,*argPLDG):
print 'Yes, pre_learn_data_grid has been run \n'
else:
print 'Running pre_learn_data_grid(hybdatadict, SDparams,prb,detectioncrit,supervised_params), \n you have not run it yet'
hash_hyb_SD,classweights,scaled_fets, target = pre_learn_data_grid(hybdatadict, SDparams,prb,detectioncrit,supervised_params)
DIRPATH = hybdatadict['output_path']
number_of_weights = len(classweights)
numspikes = scaled_fets.shape[0]
cross_valid = do_cross_validation_shuffle(numspikes,supervised_params['cross_param'])
#print cross_valid
#do_supervised(supervised_params,
#'grid_C': [1,100000,0.00001], number_cvalues = 3
number_cvalues = len(supervised_params['grid_C'])
#number_support_vectors = {}
weights_clu_test = np.zeros((number_cvalues,number_of_weights,numspikes,2),dtype=np.int32)
weights_clu_train = np.zeros((number_cvalues,number_of_weights, numspikes,2),dtype=np.int32)
cludict= {(0,0):1, (0,1):2, (1,0):3, (1,1):4}
# (prediction, groundtruth)
#(0,0) TN, (0,1) FN ,(1,0) FP ,(1,1) TP
testclu = np.zeros((number_cvalues,number_of_weights,numspikes),dtype=np.int32)
trainclu = np.zeros((number_cvalues,number_of_weights,numspikes),dtype=np.int32)
for c, Cval in enumerate(supervised_params['grid_C']):
preds = {}
preds_train = {}
##Defined to avoid: TypeError: unhashable type: 'numpy.ndarray', something about dictionaries
#testclu_pre = np.zeros((number_of_weights,numspikes),dtype=np.int32)
#trainclu_pre = np.zeros((number_of_weights,numspikes),dtype=np.int32)
for i, (weights) in enumerate(classweights):
for j, (train, test) in enumerate(cross_valid):
if supervised_params['kernel'] == 'poly':
preds[i,j], preds_train[i,j]= do_supervised_learning(test, train,Cval, supervised_params['kernel'], scaled_fets, target,classweights[i])
else:#radial kernel, only allow a single gamma value at a time
preds[i,j], preds_train[i,j]= do_supervised_learning_radial(test, train,Cval, supervised_params['kernel'],supervised_params['gamma'], scaled_fets, target,classweights[i])
print 'Computed ', classweights[i]
#Used later to make equivalent to 4 seasons clu file
weights_clu_test[c,i,test,0] = preds[i,j]
weights_clu_test[c,i,test,1] = target[test]
#Used later to make equivalent to 4 seasons clu file but for the training set
weights_clu_train[c,i,train,0] = preds_train[i,j]
weights_clu_train[c,i,train,1] = target[train]
#Make 4 seasons clu file equivalent
for k in np.arange(numspikes):
testclu[c,i,k] = cludict[tuple(weights_clu_test[c,i,k,:])]
trainclu[c,i,k] = cludict[tuple(weights_clu_train[c,i,k,:])]
# supervisedinputdict = {'test':test, 'train':train, 'Cval': Cval, 'kernel': supervised_params['kernel'], 'scaled_fets':scaled_fets, 'target', target, 'classweights': classweigths
#Add clusterings to .kwik file
supervisedparamshash = None #if addtokwik = False otherwise crashes
if addtokwik:
kwikfiles = open_files(hash_hyb_SD,dir=DIRPATH, mode='a')
supervisedparamshash = hash_utils.hash_dictionary_md5(supervised_params)
supervisedhashname = supervisedparamshash + '_' + repr(c) + '_' + repr(i)
add_clustering(kwikfiles,name = supervisedhashname + 'test', spike_clusters=testclu[c,i,:] )
add_clustering(kwikfiles,name = supervisedhashname + 'train', spike_clusters=trainclu[c,i,:] )
close_files(kwikfiles)
#print 'testclu[',c,',',i,',',k,']=',testclu[c,i,k]
# for c, Cval in enumerate(supervised_params['grid_C']):
# kwikfilename = DIRPATH + hash_hyb_SD + '.kwik'
# supervisedhashname = hash_utils.hash_dictionary_md5(detectioncrit)
# add_clustering_kwik(kwikfilename, detectedgroundtruth, detectionhashname)
####Train and test look like this for 2-fold cross validation and 200 spikes
# j = 0 train = [100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117
# 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
# 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153
# 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171
# 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189
# 190 191 192 193 194 195 196 197 198 199]
# test = [ 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
# 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
# 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74
# 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99]
# j = 1 train = [ 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
# 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
# 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74
# 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99]
# test = [100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117
# 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
# 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153
# 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171
# 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189
# 190 191 192 193 194 195 196 197 198 199]
return supervisedparamshash, classweights, testclu, trainclu
def test_detection_algorithm(hybdatadict, SDparams,prb, detectioncrit):
'''
It will query whether the cached function:
hybridata_creation_lib.create_hybrid_kwdfile(hybdatadict),
has been called already with those arguments using `joblib_utils.is_cached`,
If it has, it calls it to obtain creation_groundtruth.
else if the hybrid dataset does not exist, it will raise an Error.
creation_groundtruth consists of the equivalent to the old:
GroundtruthResfile,GroundtruthClufile,... (i.e. the times and the cluster labels for the
added hybrid spikes.
detection criteria include:
allowed_discrepancy,CSthreshold
This function will call SpikeSimilarityMeasure(a,b)
and output the file: Hash(hybdatadict)_Hash(sdparams)_Hash(detectioncrit).kwik
It will return detcrit_groundtruth, the groundtruth relative to the criteria, detectioncrit.
This will be an ordered dictionary so that hashing can work!'''
if ju.is_cached(hcl.create_hybrid_kwdfile,hybdatadict):
print 'Yes, this hybrid dataset exists, I shall now check if you have run SD \n'
meanwaveform,kwdoutputname, creation_groundtruth, amplitude = hcl.create_hybrid_kwdfile(hybdatadict)
#Take the means of the binary donor masks of the donor cluster
binmeanmask = hcl.make_average_datamask_from_mean(hybdatadict, fmask= False)
argSD = [hybdatadict,SDparams,prb]
if ju.is_cached(rsd.run_spikedetekt,*argSD):
print 'Yes, SD has been run \n'
hash_hyb_SD = rsd.run_spikedetekt(hybdatadict,SDparams,prb)
else:
print 'You need to run Spikedetekt before attempting to analyse results '
DIRPATH = hybdatadict['output_path']
with Experiment(hash_hyb_SD, dir= DIRPATH, mode='a') as expt:
res_int= expt.channel_groups[0].spikes.time_samples
res_frac = expt.channel_groups[0].spikes.time_fractional
res_int_arr = res_int[:]
res_frac_arr = res_frac[:]
detected_times = res_int_arr + res_frac_arr
#Masks
fmask = expt.channel_groups[0].spikes.features_masks
#Spikes within time window
existencewin = np.zeros_like(creation_groundtruth)
#Mean binary mask for hybrid cluster
if 'manual_meanmask' in detectioncrit.keys():
binmeanmask = detectioncrit['manual_meanmask']
else:
binmeanmask = hcl.make_average_datamask_from_mean(hybdatadict, fmask= False)
indices_in_timewindow = hash_utils.order_dictionary({})
#indices_in_timewindow = {0 (this is the 1st hybrid spike): (array([0, 1, 3]),),
#1: (array([89, 90]),),
#2: (array([154, 156, 157]),),
#3: (array([191]),),
#4: (array([259, 260, 261]),),
num_spikes_in_timewindow = hash_utils.order_dictionary({})
CauchySchwarz = hash_utils.order_dictionary({})
detected = hash_utils.order_dictionary({})
NumHybSpikes = creation_groundtruth.shape[0]
trivialmainclustering = np.zeros_like(detected_times)
detectedgroundtruth = np.zeros_like(detected_times)
print detectedgroundtruth.shape
for k in np.arange(NumHybSpikes):
list_of_differences = np.zeros((detected_times.shape[0]))
list_of_differences[:]= detected_times[:] - creation_groundtruth[k]
indices_in_timewindow[k] = np.nonzero(np.absolute(list_of_differences)<=detectioncrit['allowed_discrepancy'])
num_spikes_in_timewindow[k] = indices_in_timewindow[k][0].shape[0]
for j in np.arange(num_spikes_in_timewindow[k]):
CauchySchwarz[k,j] = SpikeSimilarityMeasure(fmask[indices_in_timewindow[k][0][j],:,1],binmeanmask[0:3*hybdatadict['numchannels']])
if CauchySchwarz[k,j] > detectioncrit['CSthreshold']:
detected[k,j] = 1
else:
detected[k,j] = 0
detectedgroundtruth[indices_in_timewindow[k][0][j]]= detected[k,j]
#Store detectedgroundtruth in a clustering
detectionhashname = hash_utils.hash_dictionary_md5(detectioncrit)
#kwikfilename = DIRPATH + hash_hyb_SD
#+ '.kwik'
#add_clustering_kwik(kwikfilename, detectedgroundtruth, detectionhashname)
kwikfiles = open_files(hash_hyb_SD,dir=DIRPATH, mode='a')
add_clustering(kwikfiles,name = detectionhashname, spike_clusters=detectedgroundtruth,overwrite = True )
print 'Added a clustering called ', detectionhashname
add_clustering(kwikfiles,name = 'main', spike_clusters= trivialmainclustering, overwrite = True)
close_files(kwikfiles)
#clusters = '/channel_groups[0]/spikes/clusters'
#detectionhash = hash_dictionary_md5(detectioncrit)
#expt.createEArray(clusters, detectionhash, tb.UInt32Atom(), (0,),
# expectedrows=1000000)
#percentage_detected = float(sum(detected.values()))/NumHybSpikes
detcrit_groundtruth_pre ={'detection_hashname':
detectionhashname,'binmeanmask': binmeanmask,'indices_in_timewindow':indices_in_timewindow, 'numspikes_in_timeswindow': num_spikes_in_timewindow,
'Cauchy_Schwarz':CauchySchwarz,'detected': detected,'detected_groundtruth': detectedgroundtruth}
detcrit_groundtruth = hash_utils.order_dictionary(detcrit_groundtruth_pre)
return detcrit_groundtruth