def main():
char2idx = {}
idx2char = []
vocabSize = 0
NN.EMBEDDINGDIM = 0
NN.NNUNITS = 0
firstSequenceToUse = ""
NN.NHIDDENLAYERS = 0
NN.HIDDENLAYERS = []
NN.TEMPERATURE = 0
NN.ReadArgsForGenerating()
with open('./NNTraining/' + NN.TRAINFILE, 'rb') as f:
char2idx = pickle.load(f)
idx2char = pickle.load(f)
vocabSize = pickle.load(f)
NN.EMBEDDINGDIM = pickle.load(f)
NN.NNUNITS = pickle.load(f)
firstSequenceToUse = pickle.load(f)
NN.NHIDDENLAYERS = pickle.load(f)
NN.HIDDENLAYERS = pickle.load(f)
NN.TEMPERATURE = pickle.load(f)
model = NN.BuildModel(vocabSize, NN.EMBEDDINGDIM, NN.NNUNITS, batchSize=1)
# model.load_weights(tf.train.latest_checkpoint(checkpointDir))
model.load_weights('./NNTraining/cp.ckpt')
model.build(tf.TensorShape([1, None]))
model.summary()
genText = NN.GenerateText(model, firstSequenceToUse, NN.WIDTH, char2idx,
idx2char)
NN.SaveFile(NN.OUTPUT, genText)
def main():
NN.ReadArgsForTrainning()
print(NN.SEQLENGTH)
if NN.DEPURATION:
print("FILES: " + str(NN.NFILES))
for f in NN.FILE:
print(" FILE: " + str(f))
print()
print("SEQUENCE LENGHT: " + str(NN.SEQLENGTH))
print()
print("BUFFER SIZE: " + str(NN.BUFFERSIZE))
print()
print("EMBEDDING DIM: " + str(NN.EMBEDDINGDIM))
print()
print("NN UNITS: " + str(NN.NNUNITS))
print()
print("EPOCHS: " + str(NN.EPOCHS))
print()
print("LAYERS: " + str(NN.NHIDDENLAYERS))
for l in NN.HIDDENLAYERS:
print(" LAYER: " + str(l))
print()
print("TEMPERATURE: " + str(NN.TEMPERATURE))
print()
print("Generating neural network:")
print()
#Auxiliar variables to store the first sequence to generate text
listDatasets = []
vocab = []
textstr = []
for f in NN.FILE:
text = NN.ReadFile(f)
v, tstr = NN.GetVocabulary(text)
vocab += (v)
textstr.append(tstr)
vocab = set(vocab)
char2idx, idx2char = NN.VectorizeText(vocab)
#print(idx2char.size)
textint = []
for t in textstr:
textint.append(NN.np.array([char2idx[s] for s in t]))
firstSeq = False
firstSequenceToUse = ""
datasets = []
examplesPerEpoch = 0
for t in textint:
print(t)
print()
firstSequence, charDataset = NN.CreateTrainingSamples(t, idx2char)
if not firstSeq:
firstSequenceToUse = firstSequence
firstSeq = True
sequencesCreated = NN.CreateSequences(NN.SEQLENGTH, charDataset)
dataset = sequencesCreated.map(NN.SplitInputTarget)
dataset = dataset.shuffle(NN.BUFFERSIZE,
False).batch(NN.GetExamplesPerEpoch(
t, NN.SEQLENGTH),
drop_remainder=True)
examplesPerEpoch = NN.GetExamplesPerEpoch(t, NN.SEQLENGTH)
datasets.append(dataset)
# Length of the vocabulary in chars
vocabSize = len(vocab)
model = NN.BuildModel(vocabSize=vocabSize,
embeddingDim=NN.EMBEDDINGDIM,
nnUnits=NN.NNUNITS,
batchSize=examplesPerEpoch)
model.summary()
model.compile(optimizer='adam', loss=NN.Loss)
# Directory where the checkpoints will be saved
checkpointDir = './NNTraining/cp.ckpt'
# Name of the checkpoint files
# checkpointPrefix = os.path.join(checkpointDir, "ckpt_{epoch}")
checkpointCallback = NN.tf.keras.callbacks.ModelCheckpoint(
filepath=checkpointDir, save_weights_only=True)
for d in datasets:
model.fit(d, epochs=NN.EPOCHS, callbacks=[checkpointCallback])
try:
os.mkdir('./NNTraining/')
except FileExistsError:
pass
trainFileName = time.strftime("%Y%m%d_%H%M%S")
with open('./NNTraining/' + trainFileName + '_Training_NN.pkl', "wb") as f:
pickle.dump(char2idx, f, pickle.HIGHEST_PROTOCOL)
pickle.dump(idx2char, f, pickle.HIGHEST_PROTOCOL)
pickle.dump(vocabSize, f, pickle.HIGHEST_PROTOCOL)
pickle.dump(NN.EMBEDDINGDIM, f, pickle.HIGHEST_PROTOCOL)
pickle.dump(NN.NNUNITS, f, pickle.HIGHEST_PROTOCOL)
pickle.dump(firstSequenceToUse, f, pickle.HIGHEST_PROTOCOL)
pickle.dump(NN.NHIDDENLAYERS, f, pickle.HIGHEST_PROTOCOL)
pickle.dump(NN.HIDDENLAYERS, f, pickle.HIGHEST_PROTOCOL)
pickle.dump(NN.TEMPERATURE, f, pickle.HIGHEST_PROTOCOL)