Skip to content

Instantly share code, notes, and snippets.


João Felipe Santos jfsantos

View GitHub Profile
from __future__ import print_function
from keras.models import Model, Sequential
from keras.layers import Input, Dense, TimeDistributed
from keras.layers.core import Reshape, Flatten, Dropout, TimeDistributedDense
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.normalization import BatchNormalization
from keras.layers.convolutional import Convolution2D
from keras.layers.recurrent import LSTM
from keras.optimizers import Adam
jfsantos / gist:0c3b19fb23ba680dbf3e
Created Dec 30, 2015
Example of classifier using Torch and the rnn module
View gist:0c3b19fb23ba680dbf3e
model = nn.Sequential()
lstm = nn.Sequencer(
lstm:remember('neither') -- force model to call forget at each call to forward
View istft.lua
signal = require "signal"
complex = require "signal.complex"
function istft(X, win, hop)
local x = torch.zeros((X:size(1)-1)*hop + win)
framesamp = X:size(2)
hopsamp = hop
for n=1,X:size(1) do
i = 1 + (n-1)*hopsamp
print(i, i + framesamp - 1)
jfsantos / error.log
Created Nov 27, 2015
char-rnn issue with new nn.Identity
View error.log
creating an lstm with 2 layers
setting forget gate biases to 1 in LSTM layer 1
setting forget gate biases to 1 in LSTM layer 2
number of parameters in the model: 240321
cloning rnn
cloning criterion
/Users/jfsantos/torch/install/bin/luajit: /Users/jfsantos/torch/install/share/lua/5.1/nn/Identity.lua:13: bad argument #1 to 'set' (expecting number or Tensor or Storage at /tmp/luarocks_torch-scm-1-8315/torch7/generic/Tensor.c:1089)
stack traceback:
[C]: in function 'set'
/Users/jfsantos/torch/install/share/lua/5.1/nn/Identity.lua:13: in function 'func'
jfsantos / gist:3dc73409d61add1243ad
Created May 24, 2015
LSTM-generated irish trad tune primed with "M: 7/8\n"
View gist:3dc73409d61add1243ad
fAed cBAG|BBBB gBeB|d2 BD E2 cB|[1cBAF G2 Bc:|[2B2 Bd e3G|
dcde fab2|afef g2 fg|affe fedB|GEDB E3 :|
jfsantos /
Last active Aug 29, 2015
Compressing the spectrum magnitude
import numpy as np
x = np.random.rand(513)
# Converting to the frequency domain
X = np.fft.rfft(x)
# Taking the log of the magnitude and converting back to rectangular
def P2R(magnitude, phase):
from collections import defaultdict
import h5py, numpy
class HDF5Matrix:
refs = defaultdict(int)
def __init__(self, datapath, dataset, start, end, normalizer=None):
if datapath not in self.refs.keys():
f = h5py.File(datapath)
self.refs[datapath] = f
import numpy as np
import scipy.signal as sig
from adaptfilt import lms
if __name__ == '__main__':
import matplotlib.pyplot as plt
from import wavfile
sigma = 0.1
order = 100
View eval_mlp.jl
ENV["MOCHA_USE_CUDA"] = "true"
using HDF5, JLD, Mocha
X = Array[]
push!(X, rand(Float32, 128,11*129,1,1))
y = Array[]
push!(y, rand(Float32, 128, 129, 1, 1))
#data_layer = AsyncHDF5DataLayer("train", "train.txt", 128, 1000, [:features, :targets], false, [])
from gammatone.fftweight import fft_gtgram
from import loadmat
s = loadmat("test.mat")["s"][:,0]
fs = 16000
# gt_py has 260 frames
gt_py = fft_gtgram(s, fs, 0.010, 0.0025, 23, 125)
# gt_mat has 269 frames