diff --git a/bob/fusion/base/algorithm/Algorithm.py b/bob/fusion/base/algorithm/Algorithm.py
index 80200e6203ccd5fc7b9d0cc8563a6fc622b538cd..621f4a4ad1899b201e5c9285fce545c4af1188e6 100644
--- a/bob/fusion/base/algorithm/Algorithm.py
+++ b/bob/fusion/base/algorithm/Algorithm.py
@@ -15,9 +15,7 @@ class Algorithm(object):
"""docstring for Algorithm"""
def __init__(self,
- scores=None,
performs_training=False,
- trainer_scores=None,
has_closed_form_solution=False,
preprocessors=None,
*args,
@@ -31,36 +29,35 @@ class Algorithm(object):
"""
super(Algorithm, self).__init__()
- self.scores = scores
self.performs_training = performs_training
- self.trainer_scores = trainer_scores
self.has_closed_form_solution = has_closed_form_solution
self.preprocessors = preprocessors
self._kwargs = kwargs
self._kwargs['preprocessors'] = preprocessors
+ def train_preprocessors(self, X):
+ if self.preprocessors is not None:
+ for preprocessor in self.preprocessors:
+ X = preprocessor.fit_transform(X)
+
def preprocess(self, scores):
if self.preprocessors is not None:
- for i, (preprocessor, trained) in enumerate(self.preprocessors):
- if not trained:
- train_scores = np.vstack(self.trainer_scores)
- preprocessor.fit(train_scores)
- self.preprocessors[i] = (preprocessor, True)
- scores = self.preprocessor.transform(scores)
+ for preprocessor in self.preprocessors:
+ scores = preprocessor.transform(scores)
return scores
- def train(self):
- negatives, positives = self.trainer_scores
- train_scores = np.vstack(self.trainer_scores)
- train_scores = self.preprocess(train_scores)
+ def train(self, train, devel=None):
+ if devel is None:
+ devel = train
+ (negatives, positives) = train
+ train_scores = np.vstack((negatives, positives))
neg_len = negatives.shape[0]
y = np.zeros((train_scores.shape[0],), dtype='bool')
y[neg_len:] = True
self.fit(train_scores, y)
- def __call__(self):
- self.scores = self.preprocess(self.scores)
- return self.decision_function(self.scores)
+ def fuse(self, scores):
+ return self.decision_function(scores)
def plot_boundary_decision(self, score_labels, threshold,
label_system1='',
diff --git a/bob/fusion/base/algorithm/MLP.py b/bob/fusion/base/algorithm/MLP.py
index 4dac5742de84080274ad41951d1c2b9b079e806b..9af9abd279653d3fbf5fd6d5d2e16140a68e74c1 100644
--- a/bob/fusion/base/algorithm/MLP.py
+++ b/bob/fusion/base/algorithm/MLP.py
@@ -5,7 +5,7 @@ from __future__ import absolute_import
import bob.learn.mlp
import bob.core.random
-
+import bob.io.base
import numpy
from .Algorithm import Algorithm
@@ -25,6 +25,8 @@ class MLP(Algorithm):
hidden_layers=None,
trainer_devel=None,
seed=None,
+ machine=None,
+ trainer=None,
*args, **kwargs):
# chicken and egg :D call __init__ twice.
super(MLP, self).__init__(performs_training=True, *args, **kwargs)
@@ -35,8 +37,11 @@ class MLP(Algorithm):
self.mlp_shape = [n_systems] + hidden_layers + [1]
super(MLP, self).__init__(
performs_training=True, mlp_shape=self.mlp_shape, seed=seed,
+ machine=str(machine), trainer=str(trainer),
*args, **kwargs)
self.seed = seed
+ self.machine = machine
+ self.trainer = trainer
self.trainer_devel = trainer_devel if trainer_devel else \
self.trainer_scores
self._my_kwargs = kwargs
@@ -85,3 +90,17 @@ class MLP(Algorithm):
if scores.ndim == 2 and scores.shape[1] == 1:
scores = scores.ravel()
return scores
+
+ def save(self, model_file):
+ d5 = bob.io.base.HDF5File(model_file, "w")
+ try:
+ self.machine.save(d5)
+ finally:
+ d5.close()
+
+ def load(self, model_file):
+ d5 = bob.io.base.HDF5File(model_file)
+ try:
+ self.machine.load(d5)
+ finally:
+ d5.close()
diff --git a/bob/fusion/base/algorithm/Weighted_Sum.py b/bob/fusion/base/algorithm/Weighted_Sum.py
index 540fcc6b9e712e7c10203c78295846b353a36af0..3f1c9ae229f2c685078085de6809dfd024e85fcc 100644
--- a/bob/fusion/base/algorithm/Weighted_Sum.py
+++ b/bob/fusion/base/algorithm/Weighted_Sum.py
@@ -20,6 +20,9 @@ class Weighted_Sum(Algorithm):
has_closed_form_solution=True, *args, **kwargs)
self.weights = weights
+ def fit(self, X, y):
+ pass
+
def decision_function(self, scores):
if self.weights is None:
return numpy.mean(scores, axis=1)
diff --git a/bob/fusion/base/algorithm/__init__.py b/bob/fusion/base/algorithm/__init__.py
index b119df913d31c817a07e39956bb3a8d9ff884601..3823cff5480839e9887344c5120688f52c64c575 100644
--- a/bob/fusion/base/algorithm/__init__.py
+++ b/bob/fusion/base/algorithm/__init__.py
@@ -2,7 +2,6 @@ from .Algorithm import Algorithm
from .Weighted_Sum import Weighted_Sum
from .LogisticRegression import LogisticRegression
from .MLP import MLP
-from .MLPClassifier import MLPClassifier
# gets sphinx autodoc done right - don't remove it
__all__ = [_ for _ in dir() if not _.startswith('_')]
diff --git a/run.sh b/run.sh
index 94c50fec6da45a7ace66e632f6395a18751f0ec0..c1cbf2135eb2efefb8b619a6a2d2a4aebdab2ef1 100755
--- a/run.sh
+++ b/run.sh
@@ -5,5 +5,5 @@ INPUT1="-i /home/amir/idiap/remote/user/mobio-male/baselines/gmm/male/ztnorm/sc
INPUT2="-I /home/amir/idiap/remote/user/mobio-male/baselines/gmm/male/ztnorm/scores-eval /home/amir/idiap/remote/user/mobio-male/baselines/gabor-graph/male/ztnorm/scores-eval /home/amir/idiap/remote/user/mobio-male/voice/gmm/male/ztnorm/scores-eval /home/amir/idiap/remote/user/mobio-male/voice/isv/male/ztnorm/scores-eval"
for HIDDEN_NODES in 5 10 25 50 100 200; do
- ./bin/fuse.py -vvv $INPUT1 $INPUT2 -o "/home/amir/idiap/remote/user/mobio-male/face-voice/F-gmm-gabor-graph-S-gmm-isv_MLP_${HIDDEN_NODES}/male/ztnorm/scores-dev" -O "/home/amir/idiap/remote/user/mobio-male/face-voice/F-gmm-gabor-graph-S-gmm-isv_MLP_${HIDDEN_NODES}/male/ztnorm/scores-eval" -a "bob.fusion.base.algorithm.MLP(preprocessors=[(sklearn.preprocessing.RobustScaler(), False)], n_systems=4, hidden_layers=[${HIDDEN_NODES}], seed=0)" --force --imports='sklearn.preprocessing' 'bob.fusion.base'
+ ./bin/fuse.py -vvv $INPUT1 $INPUT2 -o "/home/amir/idiap/remote/user/mobio-male/face-voice/F-gmm-gabor-graph-S-gmm-isv_MLP_${HIDDEN_NODES}/male/ztnorm/scores-dev" -O "/home/amir/idiap/remote/user/mobio-male/face-voice/F-gmm-gabor-graph-S-gmm-isv_MLP_${HIDDEN_NODES}/male/ztnorm/scores-eval" -a "bob.fusion.base.algorithm.MLP(preprocessors=[(sklearn.preprocessing.RobustScaler(), False)], n_systems=4, hidden_layers=[${HIDDEN_NODES}], seed=0)" --force --imports='sklearn.preprocessing','bob.fusion.base'
done