-
[Working] lstm.py Haven't been Tested yet !
-
[Working] main.py
-
[Done] Implementar Callback de Losswise en lstm.py
-
Testear modelo en lstm.py
-
Testear el main
-
[Tested] utils.py
-
[Tested] perplexity.py
-
[Tested] TokenSelector.py
-
[Not Tested] Tokenization.py
-
RNN.py
-
[Verified] Generators.py ([Thread Safe])
- Hacer lista de paquetes y crear archivo requirements.txt
-
[Done] Implementar perplexity [BPC] (bits per character)
-
Implementar perplexity per word
- [Done] Verificar que clase Generators se ajusta al uso de L y Lprima
Keras Documentation: Switching from one backend to another
cd ~/.keras/
nano keras.json
Edit keras.json, and change the field "backend" to "theano"
{
"image_data_format": "channels_last",
"epsilon": 1e-07,
"floatx": "float32",
"backend": "theano"
}
From Theano Configuration
Create a config file at home dir
cd ~
nano .theanorc
And copy and paste this configuration
[global]
device = cuda
force_device = True
floatX = float32
mode=FAST_RUN
optimizer_including=cudnn
[dnn]
enabled = True
include_path = /usr/local/cuda/include
library_path = /usr/local/cuda/lib64
[gpuarray]
preallocate = 1
[lib]
cnmem = 1
[blas]
ldflags = -lmkl_rt -lpthread
And save it.
then, on a terminal, execute
export MKL_THREADING_LAYER=GNU
and press enter
or
mañana probamos -lpthread ( ver Theano/Theano#5348 )
Edit clone_github.sh
cd ~
nano clone_github.sh
And copy and paste this script
#!/bin/bash
cd ~
cd Syllables
chmod 777 -R syllable-aware/
rm -R syllable-aware/
git clone https://github.com/nlpchile/syllable-aware.git
And save it.
conda env list
source activate base
pip install theano --upgrade
pip install keras --upgrade
source deactivate base
conda install mkl
conda install mkl-service
conda install openblas
conda install pygpu
init screen
press ctrl + a then press d
screen -r
if exists may screen this command show a list with the Ids
screen -D -r id_screen
Spanish data for tests obtained from HERE
Raw Data down obtained from Spanish Billion Words Corpus
Cristian Cardellino: Spanish Billion Words Corpus and Embeddings (March 2016), http://crscardellino.me/SBWCE/
wget http://cs.famaf.unc.edu.ar/~ccardellino/SBWCE/clean_corpus.tar.bz2
tar xf clean_corpus.tar.bz2
mv spanish_billion_words ./data/