amarion35

from keras import backend as K
from keras.engine.base_layer import Layer, InputSpec
from keras import activations
from keras.layers.convolutional import _Conv
import numpy as np

class TDNN(_Conv):
    # Original TDNN
    # A. Waibel, T. Hanazawa, G. Hinton, K. Shikano and K. J. Lang,
    # "Phoneme recognition using time-delay neural networks,"

amarion35

# Launch tensorboard
# $ tensorboard --logdir=~/tensorboard.log

import tensorflow as tf
from keras.layers import Dense, Dropout, Conv3D, MaxPooling3D, GlobalAveragePooling3D
from keras.models import Sequential
import keras.backend as K

model = Sequential()
model.add(Conv3D(8, kernel_size=(3, 3, 3), input_shape=(None,None,None,1), padding='same'))

amarion35

def plot(array, width, height):
	size = len(array)
	y = [array[int(round(k*size/width)):int(round((k+1)*size/width))] for k in range(width)]
	y = [np.mean(k) for k in y[5:]] #y[5:] when the firsts values are too big
	y = height*(y-np.min(y))/(np.max(y)-np.min(y))
	res = np.array(list([list('#'*int(round(y[k]))+' '*int(round(height-y[k]))) for k in range(len(y))]))
	res = np.rot90(res, k=1, axes=(0,1))
	for r in res:
		print(''.join(r))

amarion35

#!/usr/bin/python3
import numpy as np
import sys
import time
import math
import os

def plot(array, width, height):
        size = len(array)
        y = [array[int(round(k*size/width)):int(round((k+1)*size/width))] for k in range(width)]