Package dimer :: Package nnet

[frames] | no frames]

Package nnet

source code

abstract classes for deep models

Submodules

[hide private]

dimer.nnet.autoencoder: autoencoders
dimer.nnet.autoencoder_tests
dimer.nnet.base_test_classes: abstract tester methods and common test operations
dimer.nnet.config_spec: classes to handle configuration files
dimer.nnet.cspec_tests
dimer.nnet.monitor
dimer.nnet.monitor_tests
dimer.nnet.nccn: convolutional neural nets
dimer.nnet.nccn_tests
dimer.nnet.nnet_tests
dimer.nnet.utils: This file contains different utility functions that are not connected in anyway to the networks presented in the tutorials, but rather help in processing the outputs into a more understandable way.

Classes

[hide private]

Layer
abstract layer class.

Model
generic model class with basic functionality

SpeedLayer
This layer provides an extra set of weights as a support the momentum algorithm for SGD.

Functions

[hide private]

__spec_cls__(t)

source code

adjust_lr(err, lrmax)
adjust error rate for the next batch

source code

alloc_shared_weights_(shape, dtype, name, fan_in, rng)
alloc a matrix of weights and return a theano shared variable Note : optimal initialization of weights is dependent on the activation function used (among other things).

source code

fit_data(x)
transform each component of X so that it fits on an interval [-1, 1]

source code

scale_data(x)
transform each component of X so that it has unit variance

source code

shift_data(x)
transform each component of X so that it has zero mean

source code

verbose_compile(func)
decorator that explains what is being theano-compiled, by logging the functions __doc__

source code

Variables

[hide private]

__package__ = 'dimer.nnet'

log = <logging.Logger object at 0x10dded210>

spec_list = [<class 'dimer.nnet.config_spec.AESpec'>, <class '...

Function Details

[hide private]

alloc_shared_weights_(shape, dtype, name, fan_in, rng)

source code

alloc a matrix of weights and return a theano shared variable Note : optimal initialization of weights is dependent on the activation function used (among other things). Here I replicate http://www.deeplearning.net/tutorial/mlp.html#mlp

fit_data(x)

source code

transform each component of X so that it fits on an interval [-1, 1]

Parameters:

x - a pandas data panel of the form <anchors> X <tracks> X <genome position>

Returns:

(the scaled input, the std for each input component. this is an array of shape(<tracks>, <genome position>))

scale_data(x)

source code

transform each component of X so that it has unit variance

Parameters:

x - a pandas data panel of the form <anchors> X <tracks> X <genome position>

Returns:

(the scaled input, the std for each input component. this is an array of shape(<tracks>, <genome position>))

shift_data(x)

source code

transform each component of X so that it has zero mean

x: a pandas data panel of the form <anchors> X <tracks> X <genome position> return: (the shifted input, the mean for each input component. this is an array of shape(<tracks>, <genome position>))

Variables Details

[hide private]

spec_list

Value:

[<class 'dimer.nnet.config_spec.AESpec'>,
 <class 'dimer.nnet.config_spec.DataSpec'>,
 <class 'dimer.nnet.config_spec.ModelSpec'>,
 <class 'dimer.nnet.config_spec.MtrainSpec'>]