def main(path):
train,test,wordVocab = loaddata(path)
maxSentLen = 140
train_sent = pad_sequences(train["sent"],maxlen=maxSentLen,padding="post")
train_gen = pad_sequences(train["gen"],maxlen=5,padding="post")
train_disease = pad_sequences(train["disease"],maxlen=5,padding="post")
test_sent = pad_sequences(test["sent"],maxlen=maxSentLen,padding="post")
test_gen = pad_sequences(test["gen"],maxlen=5,padding="post")
test_disease = pad_sequences(test["disease"],maxlen=5,padding="post")
train_rel = np.array(train["rel"])
test_rel = np.array(test["rel"])
print(train_rel.shape)
print(test_rel.shape)
# load pre-embedding
print("loading embedding...")
embedding_size = 200
embeddingVocab_size = len(wordVocab)
# w2v_dir_path = "/media/network/watching_dog/embedding/bio_nlp_vec/PubMed-shuffle-win-30.bin"
w2v_dir_path = "/media/kazane/watching_dog/embedding/bio_nlp_vec/PubMed-shuffle-win-30.bin"
word2vec = KeyedVectors.load_word2vec_format(w2v_dir_path, binary=True, unicode_errors='ignore')
print("build embedding weights...")
embedding_weights = np.zeros((embeddingVocab_size + 1, embedding_size))
unknow_words = []
know_words = []
for word, index in wordVocab.items():
try:
embedding_weights[index, :] = word2vec[word.lower()]
know_words.append(word)
except KeyError as E:
# print(E)
unknow_words.append(word)
embedding_weights[index, :] = np.random.uniform(-0.025, 0.025, embedding_size)
print("unknow_per: ", len(unknow_words) / embeddingVocab_size, " unkownwords: ", len(unknow_words), " vocab_size: ",
embeddingVocab_size)
model = myModel(sent_lenth=maxSentLen,word_embedding=embedding_weights)
model.attn(embedding=embedding_weights)
model.train(inputs=[train_sent,train_gen,train_disease,],
label=[train_rel,],
save_path="./outputs/checkpoint",
validation_split=0.1,
batch_size=128,
epochs=5,
)
y_ = model.predict([test_sent,test_gen,test_disease])
# np.save(file=path+"/"+"outputs.npy",arr=y_)
is_gate = False
if is_gate is False:
y_ = get_Result(y_,)
get_F1Value(y_=y_,y=test_rel,path=path)
else:
get_ROC(y_,test_rel,path=path,gate=100)
def __init__(self):
self._view = myView(self)
self._model = myModel(self)
self._server = myServer(self)
self._view.connect('button-addChannel-clicked', self._addVideoButton)
self._view.connect('button-startChannel-clicked', self._startVideo)
self._view.connect('button-stopChannel-clicked', self._stopVideo)
self._view.connect('combobox-input-changed', self._inputChanged)
self._view.connect('destroy', self.on_destroy)
self._view.connect('button-addClient-clicked', self._addClient)
import torch
import torchvision.transforms as transforms
from torch.autograd import Variable
from model import myModel
from PIL import Image
import numpy as np
from config import Config as cfg
transform = transforms.Compose([
# transforms.Resize(128),
transforms.ToTensor()
])
dic = {0: "萝莉", 1: "御姐", 2: "熟女"}
model = myModel().eval()
model.load_state_dict(torch.load(cfg.model_saved_path))
img = Image.open(cfg.test_img)
img.show(img)
img = img.resize((128, 128))
img = transform(img).unsqueeze(0)
img = Variable(img)
out = model(img)
print(out)
pred = np.argmax(out.data.numpy())
print("这个妹子是：", dic[pred])
def _inputChanged(self, combo, channelNum, inputType):
print("input changed")
# print(combo)
print("channel", channelNum)
print("input", inputType)
self._model._setInput(channelNum, inputType)
if __name__ == "__main__":
# window = Gtk.ApplicationWindow()
view = myView()
model = myModel()
# vbox = Gtk.VBox()
# window
# window.add(vbox)
# model
controller = myController(view, model)
# Create a gstreamer pipeline with no sink.
# A sink will be created inside the GstWidget.
# widget = GstWidget('videotestsrc')
# widget.set_size_request(200, 200)
# vbox.add(widget)
unknow_words = []
know_words = []
for word, index in wordVocab.items():
try:
embedding_weights[index, :] = word2vec[word.lower()]
know_words.append(word)
except KeyError as E:
# print(E)
unknow_words.append(word)
embedding_weights[index, :] = np.random.uniform(
-0.025, 0.025, embedding_size)
print("unknow_per: ",
len(unknow_words) / embeddingVocab_size, " unkownwords: ",
len(unknow_words), " vocab_size: ", embeddingVocab_size)
print('model...')
model = myModel(sent_lenth=maxSentLen)
model.attn(embedding_weights)
# model = myModel(sent_lenth=maxSentLen)
# model.attn()
print('train...')
model.train(inputs=train, batch_size=32, epochs=5, maxSentLen=maxSentLen)
print('predict')
y_ = model.predict([test_sent, test_gene, test_phenotype])
def get_Result(y):
result = np.zeros(shape=y.shape)
for i in range(y.shape[0]):
if y[i, 0] >= y[i, 1]:
result[i, 0] = 1
else: