def train(self, dataset):
"""'Train' `CachedQueryEngine`.
Raises
------
ValueError
If `dataset`'s .fa were changed -- it would raise an
exception telling to `untrain` explicitly, since the idea is
to reuse CachedQueryEngine with the same engine and same
dataset (up to variation of .sa, such as labels permutation
"""
ds_fa_hash = idhash_(dataset.fa) + ':%d' % dataset.fa._uniform_length
if self._trained_ds_fa_hash is None:
# First time is called
self._trained_ds_fa_hash = ds_fa_hash
self._queryengine.train(dataset) # train the queryengine
self._lookup_ids = [None] * dataset.nfeatures # lookup for query_byid
self._lookup = {} # generic lookup
elif self._trained_ds_fa_hash != ds_fa_hash:
raise ValueError, \
"Feature attributes of %s (idhash=%r) were changed from " \
"what this %s was trained on (idhash=%r). Untrain it " \
"explicitly if you like to reuse it on some other data." \
% (dataset, ds_fa_hash, self, self._trained_ds_fa_hash)
else:
pass
def train(self, dataset):
"""'Train' `CachedQueryEngine`.
Raises
------
ValueError
If `dataset`'s .fa were changed -- it would raise an
exception telling to `untrain` explicitly, since the idea is
to reuse CachedQueryEngine with the same engine and same
dataset (up to variation of .sa, such as labels permutation
"""
ds_fa_hash = idhash_(dataset.fa) + ':%d' % dataset.fa._uniform_length
if self._trained_ds_fa_hash is None:
# First time is called
self._trained_ds_fa_hash = ds_fa_hash
self._qe.train(dataset) # train the qe
self._lookup_ids = [None
] * dataset.nfeatures # lookup for query_byid
self._lookup = {} # generic lookup
elif self._trained_ds_fa_hash != ds_fa_hash:
raise ValueError, \
"Feature attributes of %s (idhash=%r) were changed from " \
"what this %s was trained on (idhash=%r). Untrain it " \
"explicitly if you like to reuse it on some other data." \
% (dataset, ds_fa_hash, self, self._trained_ds_fa_hash)
else:
pass
def idhash(self):
"""To verify if dataset is in the same state as when smth else was done
Like if classifier was trained on the same dataset as in question
"""
res = 'self@%s samples@%s' % (idhash_(self), idhash_(self.samples))
for col in (self.a, self.sa, self.fa):
# We cannot count on the order the values in the dict will show up
# with `self._data.value()` and since idhash will be order-dependent
# we have to make it deterministic
keys = col.keys()
keys.sort()
for k in keys:
res += ' %s@%s' % (k, idhash_(col[k].value))
return res
def idhash(self):
"""To verify if dataset is in the same state as when smth else was done
Like if classifier was trained on the same dataset as in question
"""
res = 'self@%s samples@%s' % (idhash_(self), idhash_(self.samples))
for col in (self.a, self.sa, self.fa):
# We cannot count on the order the values in the dict will show up
# with `self._data.value()` and since idhash will be order-dependent
# we have to make it deterministic
keys = col.keys()
keys.sort()
for k in keys:
res += ' %s@%s' % (k, idhash_(col[k].value))
return res
def query(self, **kwargs):
k = idhash_(kwargs.items())
v = self._lookup.get(k, None)
if v is None:
self._lookup[k] = v = self._queryengine.query(**kwargs)
return v
def query(self, **kwargs):
k = idhash_(kwargs.items())
v = self._lookup.get(k, None)
if v is None:
self._lookup[k] = v = self._qe.query(**kwargs)
return v
