skaae

from __future__ import print_function

import gzip
import itertools
import pickle
import os
import sys

PY2 = sys.version_info[0] == 2

skaae

from __future__ import print_function

import gzip
import itertools
import pickle
import os
import sys

PY2 = sys.version_info[0] == 2

skaae

import numpy as np
import theano
import theano.tensor as T
from theano import ifelse
from .. import init
from .. import nonlinearities

from .base import Layer

skaae

def adam(loss, all_params, learning_rate=0.0002, beta1=0.1, beta2=0.001,
         epsilon=1e-8, gamma=1-1e-7):
    """
    ADAM update rules
    Default values are taken from [Kingma2014]

    References:
    [Kingma2014] Kingma, Diederik, and Jimmy Ba.
    "Adam: A Method for Stochastic Optimization."
    arXiv preprint arXiv:1412.6980 (2014).

skaae

def adam(loss, all_params, learning_rate=0.001, b1=0.9, b2=0.999, e=1e-8,
         gamma=1-1e-8):
    """
    ADAM update rules
    Default values are taken from [Kingma2014]

    References:
    [Kingma2014] Kingma, Diederik, and Jimmy Ba.
    "Adam: A Method for Stochastic Optimization."
    arXiv preprint arXiv:1412.6980 (2014).

skaae

def adam(loss, all_params, learning_rate=0.001, b1=0.9, b2=0.999, e=1e-8,
         gamma=1-1e-8):
    """
    ADAM update rules
    Default values are taken from [Kingma2014]

    References:
    [Kingma2014] Kingma, Diederik, and Jimmy Ba.
    "Adam: A Method for Stochastic Optimization."
    arXiv preprint arXiv:1412.6980 (2014).

skaae

--[[
LSTM cell. Modified from 
https://github.com/oxford-cs-ml-2015/practical6/blob/master/LSTM.lua

--]]
local LSTM = {}

-- Creates one timestep of one LSTM
function LSTM.lstm(opt)
    local x = nn.Identity()()

skaae

import numpy as np
class ConfusionMatrix:
    """
       Simple confusion matrix class
       row is the true class, column is the predicted class
    """
    def __init__(self, n_classes, class_names=None):
        self.n_classes = n_classes
        if class_names is None:
            self.class_names = map(str, range(n_classes))

skaae

lasagne.updates

lasagne.updates

The update functions implement different methods to control the learning rate for use with stochastic gradient descent.

Update functions take a loss expression or a list of gradient expressions and a list of parameters as input and return an ordered dictionary of updates:

skaae

    def get_output_for(self, input, mask=None, **kwargs):
        """
        Compute this layer's output function given a symbolic input variable

        Parameters
        ----------
        input : theano.TensorType
            Symbolic input variable.
        mask : theano.TensorType
            Theano variable denoting whether each time step in each