paddlepaddle是百度提出來的深度學習的框架，個人感覺其實和tensorflow差不多(語法上面)，因為本人也是初學者，也不是很懂tensorflow，所以，這些都是個人觀點。
百度的paddlepaddle提出貌似有一段時間了，我是最近才知道的，好奇去看了看，而且最近在看tensorflow，所以想看看paddlepaddle是不是友好一點，說實話，tensorflow還是比較難懂的(對於個人來說)。感覺paddlepaddle比tensorflow好的地方在於，paddlepaddle有百度的工程師給出對應視頻和代碼進行講解，對於入門深度學習比較好。
以下就是paddlepaddle的第一講，利用波士頓房價講解線性回歸。
模型訓練：
#-*- coding:utf-8 -*-
import sys
reload(sys)
sys.setdefaultencoding("utf-8")
import os
import paddle.v2 as paddle
import paddle.v2.dataset.uci_housing as uci_housing
with_gpu = os.getenv('WITH_GPU', '0') != '0'
def main():
# 初始化PaddlePaddle
paddle.init(use_gpu=with_gpu, trainer_count=1)
# 模型配置
x = paddle.layer.data(name='x', type=paddle.data_type.dense_vector(13))#利用前13因數來預測房價
y_predict = paddle.layer.fc(input=x, size=1, act=paddle.activation.Linear())#預測的房價值,線性激活函數
y = paddle.layer.data(name='y', type=paddle.data_type.dense_vector(1))#實際的房價值
cost = paddle.layer.square_error_cost(input=y_predict, label=y)#損失函數
# 保存網絡拓撲
inference_topology = paddle.topology.Topology(layers=y_predict)
with open("inference_topology.pkl", 'wb') as f:
inference_topology.serialize_for_inference(f)
# 創建參數
parameters = paddle.parameters.create(cost)
# 創建trainer
optimizer = paddle.optimizer.Momentum(momentum=0)#learning_rate=0.0001 學習率
trainer = paddle.trainer.SGD(
cost=cost, parameters=parameters, update_equation=optimizer)#隨機梯度下降算法
feeding = {'x': 0, 'y': 1}
# 讀取數據且打印訓練的中間信息
def event_handler(event):
if isinstance(event, paddle.event.EndIteration):
if event.batch_id % 100 == 0:
print "Pass %d, Batch %d, Cost %f" % (
event.pass_id, event.batch_id, event.cost)
if isinstance(event, paddle.event.EndPass):
if event.pass_id % 10 == 0:
with open('params_pass_%d.tar' % event.pass_id, 'w') as f:
trainer.save_parameter_to_tar(f)
result = trainer.test(
reader=paddle.batch(uci_housing.test(), batch_size=2),
feeding=feeding)#讀取房價數據，將數據打亂，每次取出2條
print "Test %d, Cost %f" % (event.pass_id, result.cost)
# 開始訓練
trainer.train(
reader=paddle.batch(
paddle.reader.shuffle(uci_housing.train(), buf_size=500),
batch_size=2),
feeding=feeding,
event_handler=event_handler,#提供一個 event_handler，來打印訓練的進度：
num_passes=30)
# 生成測試數據
test_data_creator = paddle.dataset.uci_housing.test()
test_data = []
test_label = []
#取出測試集中5條數據用於最后的預測
for item in test_data_creator():
test_data.append((item[0], ))
test_label.append(item[1])
if len(test_data) == 5:
break
#推測inference
probs = paddle.infer(
output_layer=y_predict, parameters=parameters, input=test_data)
for i in xrange(len(probs)):
print "label=" + str(test_label[i][0]) + ", predict=" + str(probs[i][0])
if __name__ == '__main__':
main()
運行結果：
Pass 0, Batch 0, Cost 886.077026 Pass 0, Batch 100, Cost 236.768433 Pass 0, Batch 200, Cost 555.669922 Test 0, Cost 56.372781 Pass 1, Batch 0, Cost 558.157104 Pass 1, Batch 100, Cost 17.486526 Pass 1, Batch 200, Cost 49.110359 Test 1, Cost 22.666769 Pass 2, Batch 0, Cost 2.017142 Pass 2, Batch 100, Cost 5.376208 Pass 2, Batch 200, Cost 1.576212 Test 2, Cost 18.296844 Pass 3, Batch 0, Cost 103.864586 Pass 3, Batch 100, Cost 84.158134 Pass 3, Batch 200, Cost 5.564497 Test 3, Cost 17.668033 Pass 4, Batch 0, Cost 2.316584 Pass 4, Batch 100, Cost 9.555552 Pass 4, Batch 200, Cost 74.418373 Test 4, Cost 17.311696 Pass 5, Batch 0, Cost 9.540855 Pass 5, Batch 100, Cost 22.676167 Pass 5, Batch 200, Cost 123.998085 Test 5, Cost 16.799527 Pass 6, Batch 0, Cost 56.558044 Pass 6, Batch 100, Cost 33.035114 Pass 6, Batch 200, Cost 58.189980 Test 6, Cost 16.333503 Pass 7, Batch 0, Cost 7.590010 Pass 7, Batch 100, Cost 34.771137 Pass 7, Batch 200, Cost 44.883244 Test 7, Cost 16.017060 Pass 8, Batch 0, Cost 42.311310 Pass 8, Batch 100, Cost 24.567163 Pass 8, Batch 200, Cost 33.340485 Test 8, Cost 15.520346 Pass 9, Batch 0, Cost 178.452744 Pass 9, Batch 100, Cost 10.791793 Pass 9, Batch 200, Cost 0.137641 Test 9, Cost 15.214742 Pass 10, Batch 0, Cost 10.072014 Pass 10, Batch 100, Cost 11.594021 Pass 10, Batch 200, Cost 24.404564 Test 10, Cost 14.916112 Pass 11, Batch 0, Cost 5.649694 Pass 11, Batch 100, Cost 31.902603 Pass 11, Batch 200, Cost 11.218608 Test 11, Cost 14.600422 Pass 12, Batch 0, Cost 87.761772 Pass 12, Batch 100, Cost 53.684475 Pass 12, Batch 200, Cost 37.861378 Test 12, Cost 14.326864 Pass 13, Batch 0, Cost 5.141076 Pass 13, Batch 100, Cost 0.324465 Pass 13, Batch 200, Cost 2.333709 Test 13, Cost 14.124264 Pass 14, Batch 0, Cost 9.482045 Pass 14, Batch 100, Cost 22.704296 Pass 14, Batch 200, Cost 12.826228 Test 14, Cost 13.945640 Pass 15, Batch 0, Cost 41.819580 Pass 15, Batch 100, Cost 10.353182 Pass 15, Batch 200, Cost 13.374403 Test 15, Cost 13.767083 Pass 16, Batch 0, Cost 83.044785 Pass 16, Batch 100, Cost 27.363625 Pass 16, Batch 200, Cost 5.347357 Test 16, Cost 13.665516 Pass 17, Batch 0, Cost 0.994224 Pass 17, Batch 100, Cost 0.298174 Pass 17, Batch 200, Cost 140.061615 Test 17, Cost 13.568394 Pass 18, Batch 0, Cost 11.832894 Pass 18, Batch 100, Cost 8.340067 Pass 18, Batch 200, Cost 30.967430 Test 18, Cost 13.465723 Pass 19, Batch 0, Cost 15.379287 Pass 19, Batch 100, Cost 123.313614 Pass 19, Batch 200, Cost 36.328705 Test 19, Cost 13.377999 Pass 20, Batch 0, Cost 12.842525 Pass 20, Batch 100, Cost 54.218903 Pass 20, Batch 200, Cost 18.377592 Test 20, Cost 13.266518 Pass 21, Batch 0, Cost 49.386784 Pass 21, Batch 100, Cost 215.253906 Pass 21, Batch 200, Cost 0.260682 Test 21, Cost 13.237288 Pass 22, Batch 0, Cost 469.974213 Pass 22, Batch 100, Cost 8.073731 Pass 22, Batch 200, Cost 0.810365 Test 22, Cost 13.192008 Pass 23, Batch 0, Cost 145.341141 Pass 23, Batch 100, Cost 15.787022 Pass 23, Batch 200, Cost 4.965213 Test 23, Cost 13.133022 Pass 24, Batch 0, Cost 10.377566 Pass 24, Batch 100, Cost 3.863908 Pass 24, Batch 200, Cost 15.857657 Test 24, Cost 13.113067 Pass 25, Batch 0, Cost 6.239013 Pass 25, Batch 100, Cost 15.914387 Pass 25, Batch 200, Cost 48.752701 Test 25, Cost 13.137239 Pass 26, Batch 0, Cost 57.843086 Pass 26, Batch 100, Cost 0.732344 Pass 26, Batch 200, Cost 48.501846 Test 26, Cost 13.141359 Pass 27, Batch 0, Cost 443.271545 Pass 27, Batch 100, Cost 227.696655 Pass 27, Batch 200, Cost 1.482114 Test 27, Cost 13.094058 Pass 28, Batch 0, Cost 11.784382 Pass 28, Batch 100, Cost 1.334578 Pass 28, Batch 200, Cost 16.487831 Test 28, Cost 13.122105 Pass 29, Batch 0, Cost 10.043719 Pass 29, Batch 100, Cost 26.890572 Pass 29, Batch 200, Cost 11.034937 Test 29, Cost 13.203439 label=8.5, predict=11.7476 label=5.0, predict=13.6822 label=11.9, predict=10.7325 label=27.9, predict=18.0696 label=17.2, predict=13.0193
房價預測：
#-*- coding:utf-8 -*- import sys reload(sys) sys.setdefaultencoding("utf-8") import paddle.v2 as paddle # Initialize PaddlePaddle. paddle.init(use_gpu=False, trainer_count=1) # Configure the neural network. x = paddle.layer.data(name='x', type=paddle.data_type.dense_vector(13)) y_predict = paddle.layer.fc(input=x, size=1, act=paddle.activation.Linear()) # Infer using provided test data. probs = paddle.infer( output_layer=y_predict, parameters=paddle.dataset.uci_housing.model(), input=[item for item in paddle.dataset.uci_housing.test()()]) for i in xrange(len(probs)): print 'Predicted price: ${:,.2f}'.format(probs[i][0] * 1000)
運行結果
Predicted price: $12,316.63 Predicted price: $13,830.34 Predicted price: $11,499.34 Predicted price: $17,395.05 Predicted price: $13,317.67 Predicted price: $16,834.08 Predicted price: $16,632.04 Predicted price: $15,384.20 Predicted price: $7,697.38 Predicted price: $13,657.83 Predicted price: $6,329.62 Predicted price: $12,153.18 Predicted price: $13,890.60 Predicted price: $11,367.41 Predicted price: $13,269.13 Predicted price: $14,979.35 Predicted price: $17,539.03 Predicted price: $16,686.41 Predicted price: $16,810.74 Predicted price: $13,620.53 Predicted price: $14,720.09 Predicted price: $12,533.42 Predicted price: $15,835.49 Predicted price: $16,064.76 Predicted price: $14,566.97 Predicted price: $13,783.11 Predicted price: $16,211.73 Predicted price: $16,362.79 Predicted price: $18,183.92 Predicted price: $16,298.03 Predicted price: $16,084.58 Predicted price: $14,406.07 Predicted price: $15,309.62 Predicted price: $12,104.60 Predicted price: $9,865.44 Predicted price: $14,116.36 Predicted price: $14,552.37 Predicted price: $16,381.32 Predicted price: $16,992.90 Predicted price: $16,722.93 Predicted price: $13,468.48 Predicted price: $13,622.97 Predicted price: $16,512.31 Predicted price: $17,004.60 Predicted price: $16,492.97 Predicted price: $16,179.70 Predicted price: $15,989.17 Predicted price: $17,289.17 Predicted price: $16,975.07 Predicted price: $18,950.22 Predicted price: $15,513.54 Predicted price: $15,652.08 Predicted price: $14,162.51 Predicted price: $14,665.31 Predicted price: $16,724.47 Predicted price: $17,369.51 Predicted price: $17,330.55 Predicted price: $17,923.71 Predicted price: $18,018.71 Predicted price: $19,392.96 Predicted price: $18,379.00 Predicted price: $17,187.61 Predicted price: $14,920.71 Predicted price: $15,435.08 Predicted price: $16,458.07 Predicted price: $17,390.93 Predicted price: $17,520.05 Predicted price: $18,763.72 Predicted price: $18,698.70 Predicted price: $20,425.67 Predicted price: $15,431.77 Predicted price: $14,803.56 Predicted price: $17,336.69 Predicted price: $13,052.34 Predicted price: $16,874.23 Predicted price: $18,547.62 Predicted price: $19,574.30 Predicted price: $21,303.89 Predicted price: $22,053.60 Predicted price: $18,862.40 Predicted price: $17,969.15 Predicted price: $19,496.96 Predicted price: $17,676.56 Predicted price: $18,699.87 Predicted price: $14,520.48 Predicted price: $12,410.05 Predicted price: $9,987.12 Predicted price: $15,381.11 Predicted price: $16,906.17 Predicted price: $21,538.57 Predicted price: $21,566.74 Predicted price: $19,905.33 Predicted price: $17,938.98 Predicted price: $20,776.08 Predicted price: $21,715.28 Predicted price: $20,169.60 Predicted price: $21,148.05 Predicted price: $22,589.09 Predicted price: $21,913.31 Predicted price: $24,388.41 Predicted price: $23,748.72 Predicted price: $22,013.94
來源：paddlepaddle官網、以上代碼對應的視頻講解地址