conf

examples/hp_grid_search.py

@@ -46,7 +46,7 @@ def grid_search():
         model.hp = hp
         # Create connected layers
         # Input layer
-        l_in = InputLayer(shape=(None, 28 * 28), name='Input')
+        l_in = InputLayer(input_shape=(None, 28 * 28), name='Input')
         # Dense Layer 1
         l_hid1 = DenseLayer(incoming=l_in, nb_units=5, W=hp.initialisation, l1=hp.l1_reg,
                             l2=hp.l2_reg, activation=hp.activation, name='Hidden layer 1')

examples/model_template.py

@@ -112,7 +112,7 @@
 
 # Create connected layers
 # Input layer
-l_in = yadll.layers.InputLayer(shape=(hp.batch_size, 28 * 28), name='Input')
+l_in = yadll.layers.InputLayer(input_shape=(hp.batch_size, 28 * 28), name='Input')
 # Dropout Layer 1
 l_dro1 = yadll.layers.Dropout(incoming=l_in, corruption_level=0.4, name='Dropout 1')
 # Dense Layer 1

examples/networks.py

@@ -33,7 +33,7 @@ def logistic_regression(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # Logistic regression Layer
     l_out = LogisticRegression(incoming=l_in, nb_class=10, name='Logistic regression')
 
@@ -59,7 +59,7 @@ def mlp(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # Dense Layer
     l_hid1 = DenseLayer(incoming=l_in, nb_units=500, W=glorot_uniform, l1=hp.l1_reg,
                         l2=hp.l2_reg, activation=tanh, name='Hidden layer 1')
@@ -92,7 +92,7 @@ def dropout(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # Dropout Layer
     l_dro1 = Dropout(incoming=l_in, corruption_level=0.4, name='Dropout 1')
     # Dense Layer
@@ -130,7 +130,7 @@ def dropconnect(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # DropConnect Layer
     l_dc1 = Dropconnect(incoming=l_in, nb_units=500, corruption_level=0.4,
                         W=glorot_uniform, activation=relu, name='Dropconnect layer 1')
@@ -165,7 +165,7 @@ def convpool(input_var=None):
     filter_shape = (20, 1, 5, 5)       # (number of filters, nb input feature maps, filter height, filter width)
     poolsize = (2, 2)                  # downsampling factor per (row, col)
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # ConvLayer needs 4D Tensor
     l_rs = ReshapeLayer(incoming=l_in, output_shape=image_shape)
     # ConvPool Layer
@@ -198,7 +198,7 @@ def lenet5(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # ConvLayer needs 4D Tensor
     image_shape = (None, 1, 28, 28)
     l_rs = ReshapeLayer(incoming=l_in, output_shape=image_shape)
@@ -251,7 +251,7 @@ def autoencoder(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # Auto Encoder Layer
     l_ae1 = AutoEncoder(incoming=l_in, nb_units=500, hyperparameters=hp_ae,
                         corruption_level=0.0, name='AutoEncoder')
@@ -285,7 +285,7 @@ def denoising_autoencoder(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # Auto Encoder Layer
     l_ae1 = AutoEncoder(incoming=l_in, nb_units=500, hyperparameters=hp_ae,
                         corruption_level=0.3, activation=relu, name='Denoising AutoEncoder')
@@ -319,7 +319,7 @@ def gaussian_denoising_autoencoder(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # Auto Encoder Layer
     l_ae1 = AutoEncoder(incoming=l_in, nb_units=500, hyperparameters=hp_ae, sigma=0.3,
                         activation=relu, name='Gaussian Denoising AutoEncoder')
@@ -353,7 +353,7 @@ def contractive_denoising_autoencoder(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # Auto Encoder Layer
     l_ae1 = AutoEncoder(incoming=l_in, nb_units=500, hyperparameters=hp_ae, contraction_level=0.3,
                         activation=relu, name='Contractive Denoising AutoEncoder')
@@ -387,7 +387,7 @@ def stacked_denoising_autoencoder(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # Auto Encoder Layer
     l_ae1 = AutoEncoder(incoming=l_in, nb_units=500, hyperparameters=hp_ae,
                         corruption_level=0.2, name='Denoising AutoEncoder 1')
@@ -425,7 +425,7 @@ def rbm(input_var=None):
 
     # Create connected layers
     # Input layer
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     # Restricted Boltzmann Machine Layer
     l_rbm1 = RBM(incoming=l_in, nb_units=500, hyperparameters=hp_ae,
                  name='Restricted Boltzmann Machine')
@@ -458,7 +458,7 @@ def dbn(input_var=None):
     hp_ae('learning_rate', 0.01)
 
     # Create connected layers
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     l_rbm1 = RBM(incoming=l_in, nb_units=500, hyperparameters=hp_ae,
                  name='Restricted Boltzmann Machine 1')
     l_rbm2 = RBM(incoming=l_rbm1, nb_units=500, hyperparameters=hp_ae,
@@ -486,7 +486,7 @@ def rnn(input_var=None):
     hp('patience', 500)
 
     # Create connected layers
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     l_rnn = RNN(incoming=l_in, n_hidden=100, n_out=28 * 28, name='Recurrent Neural Network')
     l_out = LogisticRegression(incoming=l_rnn, nb_class=10, name='Logistic regression')
 
@@ -510,7 +510,7 @@ def lstm(input_var=None):
     hp('patience', 500)
 
     # Create connected layers
-    l_in = InputLayer(shape=(None, 28 * 28), input_var=input_var, name='Input')
+    l_in = InputLayer(input_shape=(None, 28 * 28), input_var=input_var, name='Input')
     l_lstm = LSTM(incoming=l_in, n_hidden=100, n_out=28 * 28, name='Long Short Term Memory')
     l_out = LogisticRegression(incoming=l_lstm, nb_class=10, name='Logistic regression')
 

examples/updates_examples.py

@@ -34,22 +34,19 @@
 
 # Create connected layers
 # Input layer
-l_in = yadll.layers.InputLayer(shape=(hp.batch_size, 28 * 28)) #, name='Input')
+l_in = yadll.layers.InputLayer(input_shape=(hp.batch_size, 28 * 28))
 # Dropout Layer 1
-l_dro1 = yadll.layers.Dropout(incoming=l_in, corruption_level=0.5)#, name='Dropout 1')
+l_dro1 = yadll.layers.Dropout(incoming=l_in, corruption_level=0.5)
 # Dense Layer 1
 l_hid1 = yadll.layers.DenseLayer(incoming=l_dro1, nb_units=500, W=yadll.init.glorot_uniform,
-                                 l1=hp.l1_reg, l2=hp.l2_reg, activation=yadll.activation.relu) #,
-                                 #name='Hidden layer 1')
+                                 l1=hp.l1_reg, l2=hp.l2_reg, activation=yadll.activation.relu)
 # Dropout Layer 2
-l_dro2 = yadll.layers.Dropout(incoming=l_hid1, corruption_level=0.25)#, name='Dropout 2')
+l_dro2 = yadll.layers.Dropout(incoming=l_hid1, corruption_level=0.25)
 # Dense Layer 2
 l_hid2 = yadll.layers.DenseLayer(incoming=l_dro2, nb_units=500, W=yadll.init.glorot_uniform,
-                                 l1=hp.l1_reg, l2=hp.l2_reg, activation=yadll.activation.relu,
-                                 )#name='Hidden layer 2')
+                                 l1=hp.l1_reg, l2=hp.l2_reg, activation=yadll.activation.relu)
 # Logistic regression Layer
-l_out = yadll.layers.LogisticRegression(incoming=l_hid2, nb_class=10, l1=hp.l1_reg,
-                                        l2=hp.l2_reg)#, name='Logistic regression')
+l_out = yadll.layers.LogisticRegression(incoming=l_hid2, nb_class=10, l1=hp.l1_reg, l2=hp.l2_reg)
 
 # Create network and add layers
 net = yadll.network.Network('2 layers mlp with dropout')
@@ -66,6 +63,9 @@
 # updates method
 model.updates = yadll.updates.sgd
 
+# saving configuration
+conf = model.to_conf()
+
 # train the model and save it to file at each best
 model.train()
 
@@ -83,4 +83,25 @@
 print predicted_values
 print test_set_y[:30]
 
-print model
+# Reconstruction of the model from configuration
+model_2 = yadll.model.Model()
+model_2.from_conf(conf)
+
+import json
+with open('model.json', 'w') as f:
+    json.dump(model.to_json(),f)
+
+for layer in layers:
+    l = getattr(yadll.layers, 'InputLayer')(vars(model.network.layers[0]))
+
+
+class A(object):
+    def __init__(self, a, b=1, **kwargs):
+        self.a=a
+        self.b=b
+
+    def p(self):
+        print self.a
+
+
+

yadll/hyperparameters.py

@@ -74,5 +74,5 @@ def reset(self):
             self.__setattr__(name, value)
         self.iteration = 0
 
-    def to_json(self):
+    def to_conf(self):
         return self.hp_value

yadll/layers.py

@@ -33,7 +33,7 @@ class Layer(object):
     """
     nb_instances = 0
 
-    def __init__(self, incoming, name=None):
+    def __init__(self, incoming, name=None, **kwargs):
         """
         The base class that represent a single layer of any neural network.
         It has to be subclassed by any kind of layer.
@@ -107,14 +107,19 @@ def get_output(self, **kwargs):
         """
         raise NotImplementedError
 
-    def to_json(self):
-        json = {'name': self.name,
-                'input shape': self.input_shape}
-        if self.input_layer is None:
-            json['input layer'] = 'no input layer'
-        else:
-            json['input layer'] = self.input_layer.name
-        return json
+    def to_conf(self):
+        conf = self.__dict__.copy()
+        for key in ['id', 'params', 'reguls']:
+            conf.pop(key, None)
+        if conf['input_layer']:
+            conf['input_layer'] = conf['input_layer'].name
+        if 'activation' in conf:
+            conf['activation'] = conf['activation'].__name__
+        conf['type'] = self.__class__.__name__
+        return conf
+
+    def from_conf(self, conf):
+        self.__dict__.update(conf)
 
 
 class InputLayer(Layer):
@@ -125,7 +130,7 @@ class InputLayer(Layer):
     """
     nb_instances = 0
 
-    def __init__(self, shape, input_var=None, **kwargs):
+    def __init__(self, input_shape, input=None, **kwargs):
         """
         The input layer of any network
 
@@ -134,19 +139,15 @@ def __init__(self, shape, input_var=None, **kwargs):
         shape : `tuple` of `int`
             The shape of the input layer the first element is the batch size
             and can be set to None.
-        input_var : `Theano shared Variables`, optional
+        input : `Theano shared Variables`, optional
             The input data of the network, used to train the model
         """
-        super(InputLayer, self).__init__(shape, **kwargs)
-        self.input = input_var
+        super(InputLayer, self).__init__(input_shape, **kwargs)
+        self.input = input
 
     def get_output(self, **kwargs):
         return self.input
 
-    def to_json(self):
-        json = dict({'class': self.__class__.__name__}, **super(InputLayer, self).to_json())
-        return json
-
 
 class ReshapeLayer(Layer):
     """
@@ -199,6 +200,7 @@ class DenseLayer(Layer):
     def __init__(self, incoming, nb_units, W=glorot_uniform, b=constant,
                  activation=tanh, l1=None, l2=None, **kwargs):
         super(DenseLayer, self).__init__(incoming, **kwargs)
+        self.nb_units = nb_units
         self.shape = (self.input_shape[1], nb_units)
         if isinstance(W, theano.compile.SharedVariable):
             self.W = W
@@ -211,6 +213,8 @@ def __init__(self, incoming, nb_units, W=glorot_uniform, b=constant,
             self.b = initializer(b, shape=(self.shape[1],), name='b')
         self.params.append(self.b)
         self.activation = activation
+        self.l1 = l1
+        self.l2 = l2
         if l1 and l1 != 0:
             self.reguls += l1 * T.mean(T.abs_(self.W))
         if l2 and l2 != 0:
@@ -224,10 +228,6 @@ def get_output(self, **kwargs):
         X = self.input_layer.get_output(**kwargs)
         return self.activation(T.dot(X, self.W) + self.b)
 
-    def to_json(self):
-        json = dict({'class': self.__class__.__name__}, **super(DenseLayer, self).to_json())
-        return json
-
 
 class UnsupervisedLayer(DenseLayer):
     """
@@ -283,10 +283,7 @@ def __init__(self, incoming, nb_class, W=constant, activation=softmax, **kwargs)
         super(LogisticRegression, self).__init__(incoming, nb_class, W=W,
                                                  activation=activation, **kwargs)
 
-    def to_json(self):
-        json = dict({'class': self.__class__.__name__}, **super(LogisticRegression, self).to_json())
-        return json
-
+    def
 
 class Dropout(Layer):
     """
@@ -308,9 +305,6 @@ def get_output(self, stochastic=True, **kwargs):
             X = X * T_rng.binomial(self.input_shape, n=1, p=self.p, dtype=floatX)
         return X
 
-    def to_json(self):
-        json = dict({'class': self.__class__.__name__}, **super(Dropout, self).to_json())
-        return json
 
 class Dropconnect(DenseLayer):
     """

yadll/model.py

@@ -1,6 +1,6 @@
 # -*- coding: UTF-8 -*-
 import cPickle
-
+import yadll
 from .layers import *
 from .exceptions import *
 
@@ -294,11 +294,25 @@ def predict(self, X):
         predict = theano.function(inputs=[self.x], outputs=prediction, name='predict')
         return predict(X)
 
-    def to_json(self):
+    def to_conf(self):
         return {'model name': self.name,
-                'file': self.file,
-                'network': self.network.to_json(),
-                'hyperparameters': self.hp.to_json(),
-                'update': self.updates.__name__,
-                'report': self.report}
+                'hyperparameters': self.hp.to_conf(),
+                'network': self.network.to_conf(),
+                'updates': self.updates.__name__,
+                'report': self.report,
+                'file': self.file}
+
+    def from_conf(self, conf):
+        _conf = conf.copy()
+        self.name = _conf['model name']
+        self.hp = yadll.hyperparameters.Hyperparameters()
+        for k, v in _conf['hyperparameters'].iteritems():
+            self.hp(k, v)
+        self.network = yadll.network.Network()
+        self.network.from_conf(_conf['network'])
+        self.updates = getattr(yadll.updates, _conf['updates'])
+        self.report = _conf['file']
+        self.file = _conf['file']
+        pass
+