class SimilarityServer(resource.Resource):
def __init__(self):
resource.Resource.__init__(self)
self.methods = server_interface(self)
self.isLeaf = False
self.gaia = GaiaWrapper()
self.request = None
def error(self,message):
return json.dumps({'Error':message})
def getChild(self, name, request):
return self
def render_GET(self, request):
return self.methods[request.prepath[1]](request=request, **request.args)
def render_POST(self, request):
return self.methods[request.prepath[1]](request=request, **request.args)
def add_point(self, request, location, sound_id):
return json.dumps( self.gaia.add_point(location[0],sound_id[0]))
def delete_point(self, request, sound_id):
return json.dumps(self.gaia.delete_point(sound_id[0]))
def contains(self, request, sound_id):
return json.dumps(self.gaia.contains(sound_id[0]))
def get_all_point_names(self, request):
return json.dumps(self.gaia.get_all_point_names())
def get_descriptor_names(self, request):
return json.dumps({'error': False, 'result': self.gaia.descriptor_names})
def get_sounds_descriptors(self, request, sound_ids, descriptor_names=None, normalization=[0], only_leaf_descriptors=[0]):
kwargs = dict()
if descriptor_names:
kwargs['descriptor_names'] = [name for name in descriptor_names[0].split(',') if name]
kwargs['normalization'] = normalization[0] == '1'
kwargs['only_leaf_descriptors'] = only_leaf_descriptors[0] == '1'
return json.dumps(self.gaia.get_sounds_descriptors(sound_ids[0].split(','), **kwargs))
def nnsearch(self, request, sound_id=None, num_results=None, preset=None, offset=[0]):
preset = ['pca'] # For the moment we only admit pca preset, in the future we may allow more presets
if not num_results:
num_results = [DEFAULT_NUMBER_OF_RESULTS]
return json.dumps(self.gaia.search_dataset(sound_id[0], num_results[0], preset_name=preset[0], offset=offset[0]))
def api_search(self, request, target_type=None, target=None, filter=None, preset=[DEFAULT_PRESET], metric_descriptor_names=None, num_results=[DEFAULT_NUMBER_OF_RESULTS], offset=[0], in_ids=None):
'''
This function is used as an interface to all search-related gaia funcionalities we use in freesound.
This function allows the definition of a query point that will be used by gaia as the target for the search (i.e.
gaia will sort the results by similarity according to this point), and a filter to reduce the scope of the search.
The query point is defined with the target and target_type parameters, and can be either a sound id from the
gaia index, a list of descriptors with values or the content of a file attached to a post request.
Ex:
target=1234
target_type=sound_id
target=.lowlevel.pitch.mean:220
target_type=descriptors_values
target=.lowlevel.scvalleys.mean:-8.0,-11.0,-13.0,-15.0,-16.0,-15.0
target_type=descriptors_values
The filter is defined as a string following the same syntax as solr filter that is then parsed and translated to
a representation that gaia can understand.
Ex:
filter=.lowlevel.pitch.mean:[100 TO *]
TODO: how to filter multidimensional features
Along with target and filter other parameters can be specified:
- preset: set of descriptors that will be used to determine the similarity distance (by default 'lowlevel' preset)
- metric_descriptor_names*: a list of descriptor names (separated by ,) that will be used as to build the distance metric (preset will be overwritten)
* this parameter is bypassed with the descriptor_values target type, as the list is automatically built with the descriptors specified in the target
- num_results, offset: to control the number of returned results and the offset since the first result (so pagination is possible)
'''
if in_ids:
in_ids = in_ids[0].split(',')
if not target and not filter:
if target_type:
if target_type[0] != 'file':
return json.dumps({'error': True, 'result': 'At least \'target\' or \'filter\' should be specified.', 'status_code': BAD_REQUEST_CODE})
else:
return json.dumps({'error': True, 'result': 'At least \'target\' or \'filter\' should be specified.', 'status_code': BAD_REQUEST_CODE})
if target and not target_type:
return json.dumps({'error': True, 'result': 'Parameter \'target_type\' should be specified.', 'status_code': BAD_REQUEST_CODE})
if target_type:
target_type = target_type[0]
if target or target_type == 'file':
if target_type == 'sound_id':
try:
target = int(target[0])
except:
return json.dumps({'error': True, 'result': 'Invalid sound id.', 'status_code': BAD_REQUEST_CODE})
elif target_type == 'descriptor_values':
try:
target = parse_target(target[0].replace("'", '"'), self.gaia.descriptor_names['fixed-length'])
if type(target) != dict:
if type(target) == str:
return json.dumps({'error': True, 'result': target, 'status_code': BAD_REQUEST_CODE})
else:
return json.dumps({'error': True, 'result': 'Invalid descriptor values for target.', 'status_code': BAD_REQUEST_CODE})
if not target.items():
return json.dumps({'error': True, 'result': 'Invalid target.', 'status_code': BAD_REQUEST_CODE})
except Exception as e:
return json.dumps({'error': True, 'result': 'Invalid descriptor values for target.', 'status_code': BAD_REQUEST_CODE})
elif target_type == 'file':
data = request.content.getvalue().split('&')[0] # If more than one file attached, just get the first one
if not data:
return json.dumps({'error': True, 'result': 'You specified \'file\' as target file but attached no analysis file.', 'status_code': BAD_REQUEST_CODE})
try:
target = yaml.load(data)
except:
return json.dumps({'error': True, 'result': 'Analysis file could not be parsed.', 'status_code': BAD_REQUEST_CODE})
else:
return json.dumps({'error': True, 'result': 'Invalid target type.', 'status_code': BAD_REQUEST_CODE})
if filter:
try:
filter = parse_filter(filter[0].replace("'", '"'), self.gaia.descriptor_names['fixed-length'])
if type(filter) != list:
if type(filter) == str:
return json.dumps({'error': True, 'result': filter, 'status_code': BAD_REQUEST_CODE})
else:
return json.dumps({'error': True, 'result': 'Invalid filter.', 'status_code': BAD_REQUEST_CODE})
except Exception as e:
return json.dumps({'error': True, 'result': 'Invalid filter.', 'status_code': BAD_REQUEST_CODE})
'''
if preset:
if preset[0] not in PRESETS:
preset = DEFAULT_PRESET
else:
preset = preset[0]
'''
preset = 'pca' # For the moment we only admit pca preset, in the future we may allow more presets
if not num_results[0]:
num_results = int(DEFAULT_NUMBER_OF_RESULTS)
else:
num_results = int(num_results[0])
if not offset[0]:
offset = 0
else:
offset = int(offset[0])
if target_type == 'descriptor_values' and target:
metric_descriptor_names = parse_metric_descriptors(','.join(target.keys()), self.gaia.descriptor_names['fixed-length'])
else:
metric_descriptor_names = False # For the moment metric_descriptor_names can only be set by us
return json.dumps(self.gaia.api_search(target_type,
target,
filter,
preset,
metric_descriptor_names,
num_results,
offset,
in_ids))
def save(self, request, filename=None):
if not filename:
filename = [INDEX_NAME]
return json.dumps(self.gaia.save_index(filename[0]))
class SimilarityServer(resource.Resource):
def __init__(self):
resource.Resource.__init__(self)
self.methods = server_interface(self)
self.isLeaf = False
self.gaia = GaiaWrapper()
self.request = None
def error(self, message):
return json.dumps({'Error': message})
def getChild(self, name, request):
return self
def render_GET(self, request):
return self.methods[request.prepath[1]](request=request,
**request.args)
def render_POST(self, request):
return self.methods[request.prepath[1]](request=request,
**request.args)
def add_point(self, request, location, sound_id):
return json.dumps(self.gaia.add_point(location[0], sound_id[0]))
def delete_point(self, request, sound_id):
return json.dumps(self.gaia.delete_point(sound_id[0]))
def contains(self, request, sound_id):
return json.dumps(self.gaia.contains(sound_id[0]))
def get_all_point_names(self, request):
return json.dumps(self.gaia.get_all_point_names())
def get_descriptor_names(self, request):
return json.dumps({
'error': False,
'result': self.gaia.descriptor_names
})
def get_sounds_descriptors(self,
request,
sound_ids,
descriptor_names=None,
normalization=[0],
only_leaf_descriptors=[0]):
kwargs = dict()
if descriptor_names:
kwargs['descriptor_names'] = [
name for name in descriptor_names[0].split(',') if name
]
kwargs['normalization'] = normalization[0] == '1'
kwargs['only_leaf_descriptors'] = only_leaf_descriptors[0] == '1'
return json.dumps(
self.gaia.get_sounds_descriptors(sound_ids[0].split(','),
**kwargs))
def nnsearch(self,
request,
sound_id=None,
num_results=None,
preset=None,
offset=[0]):
preset = [
'pca'
] # For the moment we only admit pca preset, in the future we may allow more presets
if not num_results:
num_results = [DEFAULT_NUMBER_OF_RESULTS]
return json.dumps(
self.gaia.search_dataset(sound_id[0],
num_results[0],
preset_name=preset[0],
offset=offset[0]))
def api_search(self,
request,
target_type=None,
target=None,
filter=None,
preset=[DEFAULT_PRESET],
metric_descriptor_names=None,
num_results=[DEFAULT_NUMBER_OF_RESULTS],
offset=[0],
in_ids=None):
'''
This function is used as an interface to all search-related gaia funcionalities we use in freesound.
This function allows the definition of a query point that will be used by gaia as the target for the search (i.e.
gaia will sort the results by similarity according to this point), and a filter to reduce the scope of the search.
The query point is defined with the target and target_type parameters, and can be either a sound id from the
gaia index, a list of descriptors with values or the content of a file attached to a post request.
Ex:
target=1234
target_type=sound_id
target=.lowlevel.pitch.mean:220
target_type=descriptors_values
target=.lowlevel.scvalleys.mean:-8.0,-11.0,-13.0,-15.0,-16.0,-15.0
target_type=descriptors_values
The filter is defined as a string following the same syntax as solr filter that is then parsed and translated to
a representation that gaia can understand.
Ex:
filter=.lowlevel.pitch.mean:[100 TO *]
TODO: how to filter multidimensional features
Along with target and filter other parameters can be specified:
- preset: set of descriptors that will be used to determine the similarity distance (by default 'lowlevel' preset)
- metric_descriptor_names*: a list of descriptor names (separated by ,) that will be used as to build the distance metric (preset will be overwritten)
* this parameter is bypassed with the descriptor_values target type, as the list is automatically built with the descriptors specified in the target
- num_results, offset: to control the number of returned results and the offset since the first result (so pagination is possible)
'''
if in_ids:
in_ids = in_ids[0].split(',')
if not target and not filter:
if target_type:
if target_type[0] != 'file':
return json.dumps({
'error': True,
'result':
'At least \'target\' or \'filter\' should be specified.',
'status_code': BAD_REQUEST_CODE
})
else:
return json.dumps({
'error': True,
'result':
'At least \'target\' or \'filter\' should be specified.',
'status_code': BAD_REQUEST_CODE
})
if target and not target_type:
return json.dumps({
'error': True,
'result': 'Parameter \'target_type\' should be specified.',
'status_code': BAD_REQUEST_CODE
})
if target_type:
target_type = target_type[0]
if target or target_type == 'file':
if target_type == 'sound_id':
try:
target = int(target[0])
except:
return json.dumps({
'error': True,
'result': 'Invalid sound id.',
'status_code': BAD_REQUEST_CODE
})
elif target_type == 'descriptor_values':
try:
target = parse_target(
target[0].replace("'", '"'),
self.gaia.descriptor_names['fixed-length'])
if type(target) != dict:
if type(target) == str:
return json.dumps({
'error': True,
'result': target,
'status_code': BAD_REQUEST_CODE
})
else:
return json.dumps({
'error': True,
'result':
'Invalid descriptor values for target.',
'status_code': BAD_REQUEST_CODE
})
if not target.items():
return json.dumps({
'error': True,
'result': 'Invalid target.',
'status_code': BAD_REQUEST_CODE
})
except Exception as e:
return json.dumps({
'error': True,
'result': 'Invalid descriptor values for target.',
'status_code': BAD_REQUEST_CODE
})
elif target_type == 'file':
data = request.content.getvalue().split(
'&'
)[0] # If more than one file attached, just get the first one
if not data:
return json.dumps({
'error': True,
'result':
'You specified \'file\' as target file but attached no analysis file.',
'status_code': BAD_REQUEST_CODE
})
try:
target = yaml.load(data)
except:
return json.dumps({
'error': True,
'result': 'Analysis file could not be parsed.',
'status_code': BAD_REQUEST_CODE
})
else:
return json.dumps({
'error': True,
'result': 'Invalid target type.',
'status_code': BAD_REQUEST_CODE
})
if filter:
try:
filter = parse_filter(
filter[0].replace("'", '"'),
self.gaia.descriptor_names['fixed-length'])
if type(filter) != list:
if type(filter) == str:
return json.dumps({
'error': True,
'result': filter,
'status_code': BAD_REQUEST_CODE
})
else:
return json.dumps({
'error': True,
'result': 'Invalid filter.',
'status_code': BAD_REQUEST_CODE
})
except Exception as e:
return json.dumps({
'error': True,
'result': 'Invalid filter.',
'status_code': BAD_REQUEST_CODE
})
'''
if preset:
if preset[0] not in PRESETS:
preset = DEFAULT_PRESET
else:
preset = preset[0]
'''
preset = 'pca' # For the moment we only admit pca preset, in the future we may allow more presets
if not num_results[0]:
num_results = int(DEFAULT_NUMBER_OF_RESULTS)
else:
num_results = int(num_results[0])
if not offset[0]:
offset = 0
else:
offset = int(offset[0])
if target_type == 'descriptor_values' and target:
metric_descriptor_names = parse_metric_descriptors(
','.join(target.keys()),
self.gaia.descriptor_names['fixed-length'])
else:
metric_descriptor_names = False # For the moment metric_descriptor_names can only be set by us
return json.dumps(
self.gaia.api_search(target_type, target, filter, preset,
metric_descriptor_names, num_results, offset,
in_ids))
def save(self, request, filename=None):
if not filename:
filename = [INDEX_NAME]
return json.dumps(self.gaia.save_index(filename[0]))
class FluidSoundServer(resource.Resource):
def __init__(self):
resource.Resource.__init__(self)
self.methods = server_interface(self)
self.isLeaf = False
self.gaia = GaiaWrapper()
self.filenames = {}
if exists(settings.FILENAME_INDEX):
with open(settings.FILENAME_INDEX, "rb") as f:
self.filenames = pickle.load(f)
print(self.filenames, exists(settings.FILENAME_INDEX))
def getChild(self, name, request):
return self
def fail(self, message, status=settings.BAD_REQUEST_CODE):
return json.dumps({
'error': True,
'result': message,
'status_code': status
})
def render_GET(self, request):
return self.methods[request.prepath[1]](request=request,
**request.args)
def sound_dict(self, sound_id):
return {
"id": sound_id,
"name": self.filenames[sound_id],
"url": self.filenames[sound_id],
"username": "******",
"duration": 0 #TODO
}
############### API methods #######
def add_point(self, request, location, sound_id):
print(splitext(basename(location[0]))[0])
self.filenames[sound_id[0]] = str(splitext(basename(location[0]))[0])
print("adding " + self.filenames[sound_id[0]], sound_id)
return json.dumps(self.gaia.add_point(location[0], sound_id[0]))
def save(self, request, filename=None):
with open(settings.FILENAME_INDEX, "wb") as handle:
pickle.dump(self.filenames, handle)
filename = filename[0] if filename is not None else settings.INDEX_NAME
return json.dumps(self.gaia.save_index(filename))
def get_sound(self, request, sound_id):
if not sound_id[0] in self.filenames:
return self.fail(messages.FILE_NOT_FOUND, settings.NOT_FOUND_CODE)
else:
snd_obj = self.sound_dict(sound_id[0])
snd_obj.update({"error": False})
return json.dumps(snd_obj)
def get_descriptor_names(self, request):
return json.dumps({
'error': False,
'result': self.gaia.descriptor_names
})
def get_analysis(self,
request,
sound_ids,
descriptor_names=None,
normalization=[0],
only_leaf_descriptors=[0]):
kwargs = dict()
if descriptor_names:
kwargs['descriptor_names'] = [
name for name in descriptor_names[0].split(',') if name
]
kwargs['normalization'] = normalization[0] == '1'
kwargs['only_leaf_descriptors'] = only_leaf_descriptors[0] == '1'
return json.dumps(
self.gaia.get_sounds_descriptors(sound_ids[0].split(','),
**kwargs))
def get_similar(self,
request,
target,
filter=None,
num_results=settings.N_RESULTS):
try:
target = int(target[0])
except:
return self.fail(messages.INVALID_SOUND_ID)
r = self.api_search(request, 'sound_id', target, filter, num_results)
print(r)
return self.api_search(request, 'sound_id', target, filter,
num_results)
def get_similar_feature(self,
request,
target,
filter=None,
num_results=settings.N_RESULTS):
try:
target = parse_target(target[0].replace("'", '"'),
self.gaia.descriptor_names['fixed-length'])
if type(target) == str:
return self.fail(target)
elif type(target) != dict:
return self.fail(messages.INVALID_DESC_VALS)
if not target.items():
return self.fail(messages.INVALID_TARGET)
except Exception as e:
return self.fail(messages.INVALID_DESC_VALS)
return self.api_search(request, 'descriptor_values', target, filter,
num_results)
def api_search(self,
request,
target_type=None,
target=None,
filter=None,
num_results=settings.N_RESULTS):
if filter:
try:
result = parse_filter(
filter[0].replace("'", '"'),
self.gaia.descriptor_names['fixed-length'])
if type(result) == str:
return self.fail(result)
elif type(result) != list:
return self.fail(messages.INVALID_FILTER)
except Exception as e:
return self.fail(messages.INVALID_FILTER)
if target_type == 'descriptor_values' and target:
metric_descriptor_names = parse_metric_descriptors(
','.join(target.keys()),
self.gaia.descriptor_names['fixed-length'])
else:
metric_descriptor_names = False
result = self.gaia.api_search(
target_type,
target,
filter,
settings.DEFAULT_PRESET,
metric_descriptor_names,
num_results,
0, # TODO
None)
parsed_results = []
for r in range(len(result["result"]["results"])):
sound_id = result["result"]["results"][r][0]
if not sound_id in self.filenames:
return self.fail(messages.FILE_NOT_FOUND,
settings.NOT_FOUND_CODE)
parsed_results.append(self.sound_dict(sound_id))
result["result"]["results"] = parsed_results
return json.dumps(result)