From 33f090cb75cd40a7c1479630423e1ee87f5c2865 Mon Sep 17 00:00:00 2001
From: Tiago Freitas Pereira <tiagofrepereira@gmail.com>
Date: Sun, 28 Aug 2016 21:37:49 +0200
Subject: [PATCH] Testing the new machine concept
---
bob/learn/tensorflow/analyzers/Analizer.py | 6 +++---
.../tensorflow/network/SequenceNetwork.py | 20 ++++++++++++++++++-
.../tensorflow/script/train_mnist_siamese.py | 2 +-
.../tensorflow/trainers/SiameseTrainer.py | 3 +--
4 files changed, 24 insertions(+), 7 deletions(-)
diff --git a/bob/learn/tensorflow/analyzers/Analizer.py b/bob/learn/tensorflow/analyzers/Analizer.py
index 27ea52f0..70f54f29 100644
--- a/bob/learn/tensorflow/analyzers/Analizer.py
+++ b/bob/learn/tensorflow/analyzers/Analizer.py
@@ -21,7 +21,7 @@ class Analizer:
"""
- def __init__(self, data_shuffler, graph, feature_placeholder, session):
+ def __init__(self, data_shuffler, machine, feature_placeholder, session):
"""
Use the CNN as feature extractor for a n-class classification
@@ -33,7 +33,7 @@ class Analizer:
"""
self.data_shuffler = data_shuffler
- self.graph = graph
+ self.machine = machine
self.feature_placeholder = feature_placeholder
self.session = session
@@ -47,7 +47,7 @@ class Analizer:
data, labels = self.data_shuffler.get_batch(train_dataset=False)
feed_dict = {self.feature_placeholder: data}
- return self.session.run([self.graph], feed_dict=feed_dict)[0], labels
+ return self.machine(feed_dict, self.session)
def __call__(self):
diff --git a/bob/learn/tensorflow/network/SequenceNetwork.py b/bob/learn/tensorflow/network/SequenceNetwork.py
index 2207360f..93452b01 100644
--- a/bob/learn/tensorflow/network/SequenceNetwork.py
+++ b/bob/learn/tensorflow/network/SequenceNetwork.py
@@ -26,18 +26,29 @@ class SequenceNetwork(six.with_metaclass(abc.ABCMeta, object)):
Base constructor
**Parameters**
- input: Place Holder
+ feature_layer:
"""
self.sequence_net = OrderedDict()
self.feature_layer = feature_layer
def add(self, layer):
+ """
+ Add a layer in the sequence network
+
+ """
if not isinstance(layer, Layer):
raise ValueError("Input `layer` must be an instance of `bob.learn.tensorflow.layers.Layer`")
self.sequence_net[layer.name] = layer
def compute_graph(self, input_data, cut=False):
+ """
+ Given the current network, return the Tensorflow graph
+
+ **Parameter**
+ input_data:
+ cut:
+ """
input_offset = input_data
for k in self.sequence_net.keys():
@@ -49,3 +60,10 @@ class SequenceNetwork(six.with_metaclass(abc.ABCMeta, object)):
return input_offset
return input_offset
+
+ def compute_projection_graph(self, placeholder):
+ return self.compute_graph(placeholder, cut=True)
+
+ def __call__(self, feed_dict, session):
+ #placeholder
+ return session.run([self.graph], feed_dict=feed_dict)[0]
diff --git a/bob/learn/tensorflow/script/train_mnist_siamese.py b/bob/learn/tensorflow/script/train_mnist_siamese.py
index 42804b9f..98a83bc2 100644
--- a/bob/learn/tensorflow/script/train_mnist_siamese.py
+++ b/bob/learn/tensorflow/script/train_mnist_siamese.py
@@ -44,7 +44,7 @@ def main():
data_shuffler = PairDataShuffler(data, labels)
# Preparing the architecture
- lenet = Lenet(feature_layer="fc1")
+ lenet = Lenet(feature_layer="fc2")
loss = ContrastiveLoss()
trainer = SiameseTrainer(architecture=lenet, loss=loss, iterations=ITERATIONS, base_lr=0.00001)
diff --git a/bob/learn/tensorflow/trainers/SiameseTrainer.py b/bob/learn/tensorflow/trainers/SiameseTrainer.py
index 7b7adf5d..a6629d06 100644
--- a/bob/learn/tensorflow/trainers/SiameseTrainer.py
+++ b/bob/learn/tensorflow/trainers/SiameseTrainer.py
@@ -63,7 +63,6 @@ class SiameseTrainer(object):
train_left_graph = self.architecture.compute_graph(train_placeholder_left_data)
train_right_graph = self.architecture.compute_graph(train_placeholder_right_data)
- feature_graph = self.architecture.compute_graph(feature_placeholder, cut=True)
loss_train, within_class, between_class = self.loss(train_placeholder_labels,
train_left_graph,
@@ -88,7 +87,7 @@ class SiameseTrainer(object):
print("Initializing !!")
# Training
with tf.Session() as session:
- analizer = Analizer(data_shuffler, feature_graph, feature_placeholder, session)
+ analizer = Analizer(data_shuffler, self.architecture, feature_placeholder, session)
train_writer = tf.train.SummaryWriter('./LOGS/train',
session.graph)
--
GitLab