def fit2(): print("fit2") model = Model([ nn.Dense(128, inshape=1, activation='relu'), nn.Dense(256, activation='relu'), nn.Dropout(0.80), nn.Dense(1) ]) model.assemble() sess = Session( model, loss=losses.Mse(), optimizer=optimizers.Fixed(), ) history = sess.fit(ds, 200000, val_data=val_ds, val_epochs=1000, listeners=[ FitListener('val_end', callback=lambda h: on_val_end(sess, h)) ]) fit_report(history, report_path + '02.png', 15)
def setUpClass(cls): cls.ly_dense = nn.Dense(4, inshape=2, activation='relu') cls.ly_dropout = nn.Dropout(keep_prob=0.5) cls.ly_dense1 = nn.Dense(6) cls.ly_dropout.join(cls.ly_dense) cls.ly_dense1.join(cls.ly_dropout)
def fit_2(): model = Model( [nnlys.Dense(32, inshape=1, activation='relu'), nnlys.Dense(1)]) model.assemble() #使用了relu激活函数模型变成了非线性的, 增加了模型的复杂度 history = train(30000, ds_0, model, record_epochs=300) history['loss'] = history['loss'][5:] history['val_loss'] = history['val_loss'][5:] history['epochs'] = history['epochs'][5:] fit_report(history, report_path + '02.png')
def fit_lstm(): print("fit lstm") model = Model([ rnn.Embedding(64, vocab_size+1), wrapper.Bidirectional(rnn.LSTM(128), rnn.LSTM(128)), nn.Filter(), nn.Dense(64), nn.Dropout(0.5), nn.Dense(1, activation='linear') ]) model.assemble() fit('lstm', model)
def fit_gru(): print("fit gru") model = Model([ rnn.Embedding(64, vocab_size+1), wrapper.Bidirectional(rnn.GRU(64), rnn.GRU(64)), nn.Filter(), nn.Dense(64), nn.Dropout(0.5), nn.Dense(1, activation='linear') ]) model.assemble() fit('gru', model)
def fit_gru(): vocab_size = vocab.size() print("vocab size: ", vocab_size) model = Model([ rnn.Embedding(256, vocab_size), rnn.GRU(1024, stateful=True), nn.Dense(1024), nn.Dropout(0.5), nn.Dense(vocab_size, activation='linear') ]) model.assemble() fit("gru", model)
def fit_4(): model = Model([ nnlys.Dense(128, inshape=1, activation='relu'), nnlys.Dense(64, activation='relu'), nnlys.Dense(1) ]) model.assemble() history = train(30000, ds_1, model, batch_size=16, record_epochs=300) history['loss'] = history['loss'][20:] history['val_loss'] = history['val_loss'][20:] history['epochs'] = history['epochs'][20:] fit_report(history, report_path + '04.png')
def setUpClass(cls): cls.ly_dense = nn.Dense(4, inshape=2, activation='relu') cls.ly_dropout = nn.Dropout(keep_prob=0.5) cls.ly_dense1 = nn.Dense(6) root = RootLayer() cls.ly_dropout.set_prev(cls.ly_dense) cls.ly_dense1.set_prev(cls.ly_dropout) cls.ly_dense.set_parent(root) cls.ly_dropout.set_parent(root) cls.ly_dense1.set_parent(root) cls.ly_dense.init_params() cls.ly_dropout.init_params() cls.ly_dense1.init_params()
def fit(name, optimizer): inshape = ds_train.data.shape[1:] #pdb.set_trace() model = Model([ cnn.Conv2D(32, (3, 3), inshape=inshape), cnn.MaxPool2D((2, 2), strides=(2, 2)), cnn.Conv2D(64, (3, 3)), cnn.MaxPool2D((2, 2), strides=(2, 2)), cnn.Conv2D(64, (3, 3)), nn.Flatten(), nn.Dense(64), nn.Dropout(0.5), nn.Dense(10) ]) model.assemble() sess = Session(model, loss=losses.CategoricalCrossentropy(), optimizer=optimizer) stop_fit = session.condition_callback(lambda: sess.stop_fit(), 'val_loss', 30) accuracy = lambda h: fit_tools.accuracy(sess, ds_test, h) def save_and_report(history): #pdb.set_trace() fit_tools.fit_report(history, report_path + name + ".png") model.save(model_path + name) #pdb.set_trace() history = sess.fit(ds_train, 200, val_data=ds_test, val_steps=100, listeners=[ stop_fit, session.FitListener('val_end', callback=accuracy), session.FitListener('val_end', callback=save_and_report) ]) save_and_report(history)
def setUpClass(cls): model = Model([ cnn.Conv2D(3, (3,3), inshape=(2,12,12)), cnn.MaxPool2D(), cnn.Conv2D(4, (5, 5), padding='valid'), nn.Flatten(), nn.Dense(10) ]) model.assemble() cls.model = model
def fit(): inshape = ds_train.data.shape[1] model = Model([nn.Dense(10, inshape=inshape, activation='relu')]) model.assemble() sess = Session(model, loss=losses.CategoricalCrossentropy(), optimizer=optimizers.Fixed(0.001)) stop_fit = session.condition_callback(lambda: sess.stop_fit(), 'val_loss', 10) #pdb.set_trace() history = sess.fit(ds_train, 20000, val_epochs=5, val_data=ds_test, listeners=[ stop_fit, session.FitListener('val_end', callback=accuracy) ]) fit_report(history, report_path + "0.png")
def fit_1(): model = Model([nnlys.Dense(32, inshape=1), nnlys.Dense(1)]) model.assemble() #这个模型是一个线性模型, 用来拟合非线性函数, 模型复杂度不够,一定会表现出欠拟合 history = train(20000, ds_0, model, record_epochs=100) fit_report(history, report_path + '01.png')