EPOCHS = 1
# DIGITS = 3
# REVERSE = True
# Try replacing GRU, or SimpleRNN.
RNN = layers.LSTM
HIDDEN_SIZE = 128
BATCH_SIZE = 128
LAYERS = 1
SENTLEN = 1
EMBEDDINGS = 100
ITERATIONS = 1
sawarefData = SawarefData(MYPATH, EMBEDDINGS)
questions, expected, embeddings = sawarefData.get1DJoinedFeatures()
ctable_x = CharacterTable(set("-").union(set(questions)))
ctable_y = CharacterTable(set("-").union(set(expected)))
print(ctable_y.chars)
print(ctable_y.encode(["QAposV"], 1))
print(ctable_y.decode([0, 0], False))
# Maximum length of input is 'int + int' (e.g., '345+678'). Maximum length of
# int is DIGITS.
# MAXLEN = DIGITS + 1 + DIGITS
print('Total ayat questions:', len(questions))
print('Vectorization...')
x = np.zeros((len(questions), SENTLEN, len(ctable_x.chars) + EMBEDDINGS),
import math
from os.path import isfile
from hyperopt import Trials, tpe, STATUS_OK, fmin, hp
class colors:
ok = '\033[92m'
fail = '\033[91m'
close = '\033[0m'
#__ENCODING__
SIGNS = "+"
CHARS = "0123456789 " + SIGNS
CHAR_TABLE = CharacterTable(CHARS)
#__PARAMETERS__
PARAMETERS = {
"RNN": hp.choice("RNN", [layers.LSTM]),
"HIDDEN_SIZE": hp.choice("HIDDEN_SIZE", [64]),
"LAYERS": hp.choice("LAYERS", [1]),
"N_EPOCH": 3,
"N_BATCH": 1,
"LIMIT": 100,
"N_EXAMPLES": 10000,
"ITERATIONS": 10,
"CHAR_TABLE": CHAR_TABLE,
"CHARS": CHARS,
"SIGNS": SIGNS
}
PARAMETERS["LENGTH"] = math.ceil(
# In[31]:
# c. pad joined morphemes
e.STRING_LENGTH = max([len(x) for k in strings for x in df_strings[k]])
for s in strings:
df_strings[s] = df_strings[s].apply(padStringWithSpaces)
# In[32]:
# d. encode them in one hot encoding
charset = set("+").union(
*[list(set("".join(df_strings[x] + "-"))) for x in strings])
ctable = CharacterTable(charset, e.STRING_LENGTH)
### Now we have one shape for all strings: (STRING_LENGTH, len(charset))
for x in strings:
df_strings[x + "_onehot"] = df_strings[x].apply(ctable.encode)
df_strings['num'] = [x for x in range(len(df_strings))]
df_strings.set_index('num', append=True, inplace=True)
# In[33]:
# e. remove diac
for x in strings:
df_strings[x+ "_undiac"] = df_strings[x].apply(removeDiac)
# In[54]: