def mult_log_reg(tensor_in,
numclasses=None,
data=None,
dtype=tf.float32,
initrange=1e-10,
seed=None,
l2=0.0,
name='log_reg'):
"""
Performs mulitnomial logistic regression forward pass. Weights and bias initialized to zeros.
:param tensor_in: A tensor_ or placeholder_
:param numclasses: For classificatio
:param data: For shape inference.
:param dtype: For :any:`weights` initialization.
:param initrange: For :any:`weights` initialization.
:param seed: For :any:`weights` initialization.
:param l2: For :any:`weights` initialization.
:param name: For `variable_scope`_
:return: A tensor shape=(tensor_in.shape[0], numclasses)
"""
if data is not None:
if type(data) is loader.HotIndex:
numclasses = data.dim
elif loader.is_one_hot(data):
numclasses = data.shape[1]
else:
raise MissingShapeError('Can not infer shape from data: %s' % data)
elif numclasses is None:
raise MissingShapeError(
'Can not infer shape. Need numclasses or data argument.')
inshape = tensor_in.get_shape().as_list()
W = weights('uniform', [inshape[1], numclasses],
dtype=dtype,
initrange=initrange,
seed=seed,
l2=l2,
name=name + '_weights')
b = weights('uniform', [numclasses],
dtype=dtype,
initrange=initrange,
seed=seed,
l2=l2,
name=name + '_bias')
tensor_out = tf.nn.softmax(tf.matmul(tensor_in, W) + b)
return tensor_out
def test_is_one_hot_true_sparse():
w = sps.csr_matrix(np.array([[1, 0, 0], [1, 0, 0], [0, 1, 0], [0, 0, 1]]))
assert loader.is_one_hot(w)
def test_is_one_hot_false3_dense():
w = np.array([0, 0, 1])
assert not loader.is_one_hot(w)
def test_is_one_hot_false2_dense():
w = np.array([[0, 0, 0], [1, 0, 0], [0, 1, 0], [0, 0, 1]])
assert not loader.is_one_hot(w)
def test_is_one_hot_false1_sparse():
w = sps.csr_matrix(np.array([[5, 0, 0], [1, 0, 0], [0, 1, 0], [0, 0, 1]]))
assert not loader.is_one_hot(w)
def test_is_one_hot_true_dense():
w = np.array([[1, 0, 0], [1, 0, 0], [0, 1, 0], [0, 0, 1]])
assert loader.is_one_hot(w)