FlorianMuellerklein

mytheme.scatter = theme_bw() + theme(legend.position = 'none', panel.border = element_blank(), axis.line = element_blank(), axis.title = element_text(size = rel(1.5), face = 'bold'))

mytheme.bar = theme_bw() + theme(panel.grid.minor.x=element_blank(), panel.grid.major.x=element_blank(), axis.text.x = element_text(size = rel(1.1), face = 'bold'), axis.title.x = element_blank(), axis.text.y = element_text(size = rel(1.1), face = 'bold'), axis.title.y = element_text(vjust = 1), legend.position = 'none', panel.border = element_blank(), axis.line = element_blank(), axis.title = element_text(size = rel(1.5), face = 'bold'))

FlorianMuellerklein

import pandas as pd
import numpy as np

from sklearn.ensemble import GradientBoostingClassifier

train_loc = 'train.csv'
test_loc = 'test.csv'
TREES = 30
NODES = 7

FlorianMuellerklein

import lasagne
from lasagne.nonlinearities import rectify, softmax
from lasagne.layers import InputLayer, DenseLayer, DropoutLayer, batch_norm, BatchNormLayer
from lasagne.layers import ElemwiseSumLayer, NonlinearityLayer, GlobalPoolLayer
from lasagne.layers.dnn import Conv2DDNNLayer as ConvLayer
from lasagne.init import HeNormal

def ResNet_FullPre_Wide(input_var=None, n=3, k=2):
    '''
    Adapted from https://github.com/Lasagne/Recipes/tree/master/papers/deep_residual_learning.

FlorianMuellerklein

#!/usr/bin/env python

"""
Lasagne example of ResNet-110 from 'Identity Mappings in Deep Residual Networks', Kaiming He et al. 2016 (https://arxiv.org/abs/1603.05027)
"""

from __future__ import print_function

import sys

FlorianMuellerklein

import multiprocessing as mp

class threaded_batch_iter(object):
    '''
    Batch iterator to make transformations on the data. 
    Uses multiprocessing so that batches can be created on CPU while GPU runs previous batch
    '''
    def __init__(self, batchsize):
        self.batchsize = batchsize

FlorianMuellerklein

from keras.layers import merge, Input, Dropout
from keras.layers import Dense, Activation
from keras.layers import Convolution2D, MaxPooling2D
from keras.layers.pooling import GlobalAveragePooling2D
from keras.layers import BatchNormalization
from keras.models import Model

'''
Adapted from https://arxiv.org/pdf/1611.10080v1.pdf.
It's a wide and deep residual network designed for optimal feature extraction and gradient flow.

FlorianMuellerklein

FlorianMuellerklein

This code snippet shows how we can change a layer in a pretrained model. In the following code, we change all the ReLU activation functions with SELU in a resnet18 model.

import torch
from torchvision import model

resnet18 = model.resnet18(pretrained=True)

def funct(list_mods):