Logistische Regression mit mehreren Klassen

Question

Ich versuche, die logistische Regression für ein Beispiel für ein Multi-Class zu verstehen, und ich habe den folgenden Code:

import numpy
import theano
import theano.tensor as T
rng = numpy.random

num_classes = 3
#N = number of examples
N = 100
#feats = number of input neurons
feats = 784
#training rate
tr_rate = 0.1 

D = (rng.randn(N, feats), rng.randint(size=N, low=0, high=num_classes))
training_steps = 1000

# Declare Theano symbolic variables
x = T.matrix("x")
y = T.ivector("y")

w = theano.shared(rng.randn(feats), name="w")
b = theano.shared(0.01, name="b")

# Construct Theano expression graph
sigma = T.nnet.softmax(T.dot(x,w) + b) 
prediction = T.argmax(sigma, axis=1)          # The class with highest probability

# Cross-entropy loss function
xent = -T.mean(T.log(sigma)[T.arange(y.shape[0]), y])

cost = xent.mean() + 0.01 * (w ** 2).sum()      # Regularisation 
gw, gb = T.grad(cost, [w, b])                   # Compute the gradient of the cost 

# Compile
train = theano.function(
          inputs=[x,y],
          outputs=[xent],
          updates=((w, w - tr_rate * gw), (b, b - tr_rate * gb)),
          allow_input_downcast=True)
predict = theano.function(inputs=[x], outputs=prediction)

# Train
for i in range(training_steps):
    train(D[0], D[1])      

Zu diesem Zeitpunkt gibt mir der Code jedoch den folgenden Fehler:

ValueError                                Traceback (most recent call last)
<ipython-input-246-c4753ce8ccc7> in <module>()
      1 # Train
      2 for i in range(training_steps):
----> 3     train(D[0], D[1])

/Users/vzocca/gitProjects/Theano/theano/compile/function_module.pyc in __call__(self, *args, **kwargs)
    869                     node=self.fn.nodes[self.fn.position_of_error],
    870                     thunk=thunk,
--> 871                     storage_map=getattr(self.fn, 'storage_map', None))
    872             else:
    873                 # old-style linkers raise their own exceptions

/Users/vzocca/gitProjects/Theano/theano/gof/link.pyc in raise_with_op(node, thunk, exc_info, storage_map)
    312         # extra long error message in that case.
    313         pass
--> 314     reraise(exc_type, exc_value, exc_trace)
    315 
    316 

/Users/vzocca/gitProjects/Theano/theano/compile/function_module.pyc in __call__(self, *args, **kwargs)
    857         t0_fn = time.time()
    858         try:
--> 859             outputs = self.fn()
    860         except Exception:
    861             if hasattr(self.fn, 'position_of_error'):

ValueError: number of rows in x (1) does not match length of y (100)
Apply node that caused the error: CrossentropySoftmaxArgmax1HotWithBias(Elemwise{Add}[(0, 0)].0, Alloc.0, y)
Toposort index: 12
Inputs types: [TensorType(float64, row), TensorType(float64, vector), TensorType(int32, vector)]
Inputs shapes: [(1, 100), (100,), (100,)]
Inputs strides: [(800, 8), (8,), (4,)]
Inputs values: ['not shown', 'not shown', 'not shown']
Outputs clients: [[Sum{acc_dtype=float64}(CrossentropySoftmaxArgmax1HotWithBias.0)], [], []]

Backtrace when the node is created:
  File "<ipython-input-244-834f04b2fb36>", line 16, in <module>
    xent = -T.mean(T.log(sigma)[T.arange(y.shape[0]), y])

Hinweis: Verwenden Sie das Theano Flag 'Exception_verbosity = High' für einen Debugprint und eine Speicherkarte dieses Anwendungsknotens.

Kann jemand helfen zu verstehen, was falsch ist? Vielen Dank!

Answer 1

Ich fand meine eigene Antwort. Ich hatte definiert

w = theano.shared(rng.randn(feats), name="w")

Und das war falsch. Die richtige Definition ist:

w = theano.shared(rng.randn(feats, num_classes), name="w")

Da die Gewichte Link 'feats'-zum des Eingangsneurons zu' num_classes'-notzahl von Ausgangsneuronen.