# coding: utf-8
  
  # In[29]:
  
  # Import
  import itertools
  import shelve
  import pickle
  import pandas
  import numpy
  import nltk
  import codecs
  import gensim
  import scipy
  from scipy import sparse
  import scipy.sparse
  import scipy.io
  import sklearn
  from sklearn.feature_extraction.text import CountVectorizer
  import sklearn.metrics
  from sklearn.neighbors import NearestNeighbors
  from sklearn.metrics import confusion_matrix
  from sklearn import preprocessing
  from keras.models import Sequential
  from keras.layers.core import Dense, Dropout, Activation,AutoEncoder
  from keras.optimizers import SGD,Adam
  from keras.layers import containers
  from mlp import *
  import mlp
  import sys
  import utils
  
  
  
  
  # In[3]:
  
  
  
  # In[4]:
  
  #db=shelve.open("SPELIKE_MLP_DB.shelve",writeback=True)
  corps=shelve.open(sys.argv[1]+".shelve")
  ##
  #
  
  out_db=shelve.open(sys.argv[2]+".shelve",writeback=True)
  
  out_db_do=shelve.open(sys.argv[2]+"_DOMLP.shelve",writeback=True)
  
  lb=preprocessing.LabelBinarizer()
  y_train=lb.fit_transform([utils.select(ligneid) for ligneid in corps["LABEL"]["TRAIN"]])
  y_dev=lb.transform([utils.select(ligneid) for ligneid in corps["LABEL"]["DEV"]])
  y_test=lb.transform([utils.select(ligneid) for ligneid in corps["LABEL"]["TEST"]])
  
  
  keys = corps.keys()
  if "LABEL" in keys :
      keys.remove("LABEL")
  for key in keys:
      print key
      try:
          x_train=corps[key]["TRAIN"].todense()
          x_dev=corps[key]["DEV"].todense()
          x_test=corps[key]["TEST"].todense()
      except :
          x_train=corps[key]["TRAIN"]
          x_dev=corps[key]["DEV"]
          x_test=corps[key]["TEST"]
  
      out_db[key]=mlp.train_mlp(x_train,y_train,x_dev,y_dev,x_test,y_test,[256,128,256],dropouts=[0,0,0],sgd=Adam(lr=0.0001),epochs=150)
      out_db_do[key]=mlp.train_mlp(x_train,y_train,x_dev,y_dev,x_test,y_test,[256,128,256],dropouts=[0.5,0,0],sgd=Adam(lr=0.0001),epochs=150)
  
  
  corps.close()
  out_db.close()