def put_in_split(split,image_config_gen,m,task,ext_hash,split_id,split_fs):
out_record = SON([('model',m['config']['model']),
('images',son_escape(image_config_gen['images'])),
('task',son_escape(task)),
('split_id',split_id),
])
filename = get_filename(out_record)
out_record['filename'] = filename
out_record['__hash__'] = ext_hash
print('dump out ...')
out_data = cPickle.dumps(SON([('split',split)]))
split_fs.put(out_data,**out_record)
def train_test_loop(outfile,extract_creates,task_config,feature_config_path,hash):
feature_config = get_config(feature_config_path)
base_query = SON([('__config_hash__',hash)])
image_params = SON([('image',feature_config['image'])])
models_params = feature_config['models']
ntrain = task_config['ntrain']
ntest = task_config['ntest']
ntrain_pos = task_config.get('ntrain_pos')
N = task_config.get('N',10)
query = task_config['query']
base_query.update(reach_in('config',task_config.get('universe',SON([]))))
print('\n')
print('BASE',base_query)
print('\n')
conn = pm.Connection(document_class=SON)
db = conn['v1']
fs = gridfs.GridFS(db, collection = 'model_performance')
cquery = reach_in('config',query)
for m in models_params:
base_query_copy = base_query.copy()
base_query_copy.update(reach_in('config.model',m))
splitdata, results = train_test(cquery,'v1','features',ntrain,ntest,ntrain_pos=ntrain_pos,N=N,universe=base_query_copy)
splitpickle = cPickle.dumps(splitdata)
data = SON([('feature_config_path',feature_config_path),
('model',m),
('task',son_escape(task_config)),
('image__aggregate__',son_escape(feature_config['image']))])
filename = get_filename(data)
data.update(results)
data['filename'] = filename
fs.put(splitpickle,**data)
createCertificateDict(outfile,{'task_config':task_config,'feature_config':feature_config,'feature_config_path':feature_config_path})
def put_in_split_result(res,image_config_gen,m,task,ext_hash,split_id,splitres_fs):
out_record = SON([('model',m['config']['model']),
('images',son_escape(image_config_gen['images'])),
('task',son_escape(task)),
('split_id',split_id),
])
split_result = SON([])
for stat in STATS:
if stat in res and res[stat] != None:
split_result[stat] = res[stat]
filename = get_filename(out_record)
out_record['filename'] = filename
out_record['__hash__'] = ext_hash
out_record.update(split_result)
print('dump out split result...')
out_data = cPickle.dumps(SON([('split_result',res)]))
splitres_fs.put(out_data,**out_record)
def put_in_performance(split_results,image_config_gen,m,model_hash,image_hash,perf_coll,task,ext_hash):
model_results = SON([])
for stat in STATS:
if stat in split_results[0] and split_results[0][stat] != None:
model_results[stat] = sp.array([split_result[stat] for split_result in split_results]).mean()
out_record = SON([('model',m['config']['model']),
('model_hash',model_hash),
('model_filename',m['filename']),
('images',son_escape(image_config_gen['images'])),
('image_hash',image_hash),
('task',son_escape(task)),
('__hash__',ext_hash)
])
out_record.update(model_results)
perf_coll.insert(out_record)
def greedy_optimization(outfile,task,image_certificate_file,initial_model,convolve_func,rep_limit, modifier_args,modifier):
conn = pm.Connection(document_class=bson.SON)
db = conn['v1']
opt_fs = gridfs.GridFS(db,'optimized_performance')
image_coll = db['raw_images.files']
image_fs = gridfs.GridFS(db,'raw_images')
image_certdict = cPickle.load(open(image_certificate_file))
print('using image certificate', image_certificate_file)
image_hash = image_certdict['run_hash']
image_args = image_certdict['out_args']
if convolve_func == v1f.v1like_filter_pyfft:
v1_pyfft.setup_pyfft()
filterbanks = []
perfs = []
model_configs = []
center_config = initial_model
i = 0
improving = True
while ((i < rep_limit) or rep_limit is None):
i += 1
print('Round', i)
next_configs = [m for m in get_consistent_deltas(center_config,modifier) if m not in model_configs]
if next_configs:
next_results = [get_performance(task,image_hash,image_fs,m,convolve_func) for m in next_configs]
next_perfs = [x[0] for x in next_results]
next_filterbanks = [x[1] for x in next_results]
next_perf_ac_max = np.array([x['test_accuracy'] for x in next_perfs]).max()
perf_ac_max = max([x['test_accuracy'] for x in perfs]) if perfs else 0
if next_perf_ac_max > perf_ac_max:
next_perf_ac_argmax = np.array([x['test_accuracy'] for x in next_perfs]).argmax()
center_config = next_configs[next_perf_ac_argmax]
print('\n\n')
print('new best performance is', next_perf_ac_max, 'from model', center_config)
print('\n\n')
perfs.extend(next_perfs)
model_configs.extend(next_configs)
filterbanks.extend(next_filterbanks)
else:
print('Breaking because no further optimization could be done. Best existing performance was', perf_ac_max, 'while best next performance was', next_perf_ac_max)
break
else:
print('Breaking because no next configs')
break
perfargmax = np.array([p['test_accuracy'] for p in perfs]).argmax()
best_model = model_configs[perfargmax]
best_performance = perfs[perfargmax]
out_record = SON([('initial_model',initial_model),
('task',son_escape(task)),
('images',son_escape(image_args)),
('images_hash',image_hash),
('modifier_args',son_escape(modifier_args)),
('modifier',modifier.__class__.__module__ + '.' + modifier.__class__.__name__)
])
filename = get_filename(out_record)
out_record['filename'] = filename
out_record.update(SON([('performances',perfs)]))
out_record.update(SON([('best_model',best_model)]))
out_record.update(SON([('best_performance',best_performance)]))
out_record.update(SON([('num_steps',len(model_configs))]))
out_record.update(SON([('models',model_configs)]))
outdata = cPickle.dumps(filterbanks)
opt_fs.put(outdata,**out_record)
if convolve_func == v1f.v1like_filter_pyfft:
v1_pyfft.cleanup_pyfft()
createCertificateDict(outfile,{'image_file':image_certificate_file})
""" Example Parameters module
"""
#from collections import OrderedDict
import copy
import itertools
from bson import SON
import config.ten_categories_images as Images
import config.renderman_correlation_tasks2 as Tasks
from dbutils import son_escape
import config.ht_l1_gabor_models_for_corr as l1_models
corr_layer = SON([(u'filter',SON([(u'model_name','correlation'),
(u'random_subset',SON([('const',.5)])),
(u'images',son_escape(Images.config['images']))])),
(u'activ', SON([(u'min_out', 0),
(u'max_out', 1)]))])
models = []
for M in l1_models.config['models']:
for num_filters in [256,384]:
for task in Tasks.config['extractions']:
m = copy.deepcopy(M)
m['layers'].append(copy.deepcopy(corr_layer))
m['layers'][2]['filter']['task'] = son_escape(task)
m['layers'][2]['filter']['num_filters'] = num_filters
models.append(m)
(u'ker_shape', [13, 13]),
(u'divfreqs', [2, 4, 7, 8, 11]),
(u'norients', 7)])),
(u'activ', SON([(u'min_out', 0),
(u'max_out', 1)])),
(u'lnorm', SON([(u'inker_shape', [9, 9]),
(u'outker_shape', [9, 9]),
(u'threshold', 10.0),
(u'stretch', 0.1)])),
(u'lpool', SON([(u'order', 2),
(u'stride', 2),
(u'ker_shape', [5, 5])]))]),
SON([(u'filter',SON([(u'model_name','correlation'),
(u'num_filters',256),
(u'random_subset',SON([('const',.5)])),
(u'task',son_escape(Tasks.config['extractions'][0])),
(u'images',son_escape(Images.config['images']))])),
(u'activ', SON([(u'min_out', 0),
(u'max_out', 1)]))])
]),
])
config = {
'models': [model]
}
(u"stretch", 0.1),
]
),
),
(u"lpool", SON([(u"order", 2), (u"stride", 2), (u"ker_shape", [5, 5])])),
]
),
SON(
[
(
u"filter",
SON(
[
(u"model_name", "correlation"),
(u"random_subset", SON([("const", 0.5)])),
(u"num_filters", 256),
(u"task", son_escape(Tasks.config["extractions"][0])),
(u"images", son_escape(Images.config["images"])),
]
),
),
(u"activ", SON([(u"min_out", 0), (u"max_out", 1)])),
]
),
],
),
]
)
config = {"models": [model]}
def extract_and_evaluate_parallel(outfile,image_certificate_file,model_certificate_file,cpath,convolve_func_name,task,ext_hash):
(model_configs, image_config_gen, model_hash, image_hash, task_list,
perf_col, split_coll, split_fs, splitperf_coll, splitperf_fs) = prepare_extract_and_evaluate(ext_hash,
image_certificate_file,
model_certificate_file,
task)
jobids = []
if convolve_func_name == 'numpy':
opstring = '-l qname=extraction_cpu.q'
elif convolve_func_name == 'pyfft':
opstring = '-l qname=extraction_gpu.q -o /home/render -e /home/render'
for m in model_configs:
print('Evaluating model',m)
for task in task_list:
classifier_kwargs = task.get('classifier_kwargs',{})
print('task',task)
splits = generate_splits(task,image_hash,'images')
for (ind,split) in enumerate(splits):
put_in_split(split,image_config_gen,m,task,ext_hash,ind,split_fs)
jobid = qsub(extract_and_evaluate_parallel_core,(image_config_gen,m,task,ext_hash,ind,convolve_func_name),opstring=opstring)
jobids.append(jobid)
print(jobids)
statuses = wait_and_get_statuses(jobids)
for m in model_configs:
print('Evaluating model',m)
for task in task_list:
split_results = get_most_recent_files(splitperf_coll,{'__hash__':ext_hash,'task':son_escape(task),'model':m['config']['model'],'images':son_escape(image_config_gen['images'])})
put_in_performance(split_results,image_config_gen,m,model_hash,image_hash,perf_col,task,ext_hash)
createCertificateDict(outfile,{'image_file':image_certificate_file,'models_file':model_certificate_file})
def extract_and_evaluate_parallel_core(image_config_gen,m,task,ext_hash,split_id,convolve_func_name,cache_port=None):
if cache_port is None:
cache_port = NETWORK_CACHE_PORT
conn = pm.Connection(document_class=bson.SON)
db = conn[DB_NAME]
split_col = db['splits.files']
split_fs = gridfs.GridFS(db,'splits')
splitconf = get_most_recent_files(split_col,{'__hash__':ext_hash,'split_id':split_id,'model':m['config']['model'],'images':son_escape(image_config_gen['images'])})[0]
split = cPickle.loads(split_fs.get_version(splitconf['filename']).read())['split']
res = extract_and_evaluate_core(split,m,convolve_func_name,task,cache_port)
splitperf_fs = gridfs.GridFS(db,'split_performance')
put_in_split_result(res,image_config_gen,m,task,ext_hash,split_id,splitperf_fs)
def evaluate(outfile,feature_certificate,cpath,task,ext_hash):
conn = pm.Connection(document_class=bson.SON)
db = conn[DB_NAME]
perf_fs = gridfs.GridFS(db,'performance')
perf_coll = db['performance.files']
remove_existing(perf_coll,perf_fs,ext_hash)
feature_certdict = cPickle.load(open(feature_certificate))
feature_hash = feature_certdict['feature_hash']
image_hash = feature_certdict['image_hash']
model_hash = feature_certdict['model_hash']
image_config_gen = feature_certdict['args']['images']
model_col = db['models.files']
feature_fs = gridfs.GridFS(db,'features')
feature_col = db['features.files']
stats = ['test_accuracy','ap','auc','mean_ap','mean_auc','train_accuracy']
if isinstance(task,list):
task_list = task
else:
task_list = [task]
model_configs = get_most_recent_files(model_col,{'__hash__':model_hash})
for m in model_configs:
print('Evaluating model',m)
for task in task_list:
task['universe'] = task.get('universe',SON([]))
task['universe']['model'] = m['config']['model']
print('task', task)
classifier_kwargs = task.get('classifier_kwargs',{})
split_results = []
splits = generate_splits(task,feature_hash,'features')
for (ind,split) in enumerate(splits):
print ('split', ind)
train_data = split['train_data']
test_data = split['test_data']
train_filenames = [t['filename'] for t in train_data]
test_filenames = [t['filename'] for t in test_data]
assert set(train_filenames).intersection(test_filenames) == set([])
print('train feature extraction ...')
train_features = sp.row_stack([load_features(f['filename'],feature_fs,m,task) for f in train_data])
print('test feature extraction ...')
test_features = sp.row_stack([load_features(f['filename'],feature_fs,m,task) for f in test_data])
train_labels = split['train_labels']
test_labels = split['test_labels']
print('classifier ...')
res = svm.classify(train_features,train_labels,test_features,test_labels,classifier_kwargs)
print('Split test accuracy', res['test_accuracy'])
split_results.append(res)
model_results = SON([])
for stat in STATS:
if stat in split_results[0] and split_results[0][stat] != None:
model_results[stat] = sp.array([split_result[stat] for split_result in split_results]).mean()
out_record = SON([('model',m['config']['model']),
('model_hash',model_hash),
('model_filename',m['filename']),
('images',son_escape(image_config_gen)),
('image_hash',image_hash),
('task',son_escape(task)),
])
filename = get_filename(out_record)
out_record['filename'] = filename
out_record['config_path'] = cpath
out_record['__hash__'] = ext_hash
out_record.update(model_results)
print('dump out ...')
out_data = cPickle.dumps(SON([('split_results',split_results),('splits',splits)]))
perf_fs.put(out_data,**out_record)
createCertificateDict(outfile,{'feature_file':feature_certificate})