一. 数据集的准备与预处理
1 . 收集dataset
(大量用户名--包含正常用户名与非法用户名)
包含两个txt文件 legal_name.txt ilegal_name.txt. 如下图所示
2. 用文件进行预处理
# Data setsimport osimport pandas as pdDATAPATH = "../dataset"POS = os.path.join(DATAPATH, "legal_name.txt")POS_OUTPUT = os.path.join(DATAPATH, "legal_name.csv")NEG = os.path.join(DATAPATH, "ilegal_name.txt")NEG_OUTPUT = os.path.join(DATAPATH, "ilegal_name.csv")def process_org_data(input_data, output_data, lable):reader = pd.read_csv(input_data, iterator=True)while True:try:train = reader.get_chunk(10000)train['username'] = train['username'].astype(str)train['username'] = map(lambda x: x.strip(), train['username'])train['length'] = train['username'].apply(len)... ...train['label'] = map(lambda x: lable, train['username'])train.to_csv(output_data, encoding='utf-8', mode='a', index=False, header=False)except StopIteration:print "Iteration is stopped."breakif __name__ == '__main__':process_org_data(POS, POS_OUTPUT, 1)process_org_data(NEG, NEG_OUTPUT, 0)
根据需求提取相应的特征, 输出成 csv 格式,包含特征列与label列
把合法用户dataset与非法用户dataset,合并打乱,切割成 train.csv 和 test.csv
pos_dataset = read_dataset(POS)neg_dataset = read_dataset(NEG)dataset = pd.concat([pos_dataset, neg_dataset])dataset = dataset.sample(frac=1).reset_index(drop=True)train_data = dataset.loc[:200000, :]test_data = dataset.loc[200000:, :]train_data.to_csv(os.path.join(DataPath, "train.csv"), index=False)test_data.to_csv(os.path.join(DataPath, "test.csv"), index=False)
二. Keras 构建DNN模型进行训练与模型保存
import pandas as pdimport osimport tensorflow as tfimport matplotlib.pyplot as pltfrom sklearn.metrics import confusion_matrixfrom sklearn.metrics import classification_reportimport numpy as nptf.logging.set_verbosity(tf.logging.INFO)DataPath = "../dataset"TRAIN = os.path.join(DataPath, "train.csv")TEST = os.path.join(DataPath, "test.csv")COLUMNS = ["username", ... , "label"]train_dataset = pd.read_csv(TRAIN, skipinitialspace=True, skiprows=1, names=COLUMNS)test_dataset = pd.read_csv(TEST, skipinitialspace=True, skiprows=1, names=COLUMNS)for col in train_dataset.columns[1:]:train_dataset[col] = pd.to_numeric(train_dataset[col], errors='coerce')for col in test_dataset.columns[1:]:test_dataset[col] = pd.to_numeric(test_dataset[col], errors='coerce')X_train = train_dataset.iloc[:, range(1, 19)].valuesy_train = train_dataset.iloc[:, 19].valuesX_test = test_dataset.iloc[:, range(1, 19)].valuesy_test = test_dataset.iloc[:, 19].valuesdef build_model():############model = tf.keras.models.Sequential()model.add(tf.keras.layers.Dense(64, input_dim=18))# model.add(tf.keras.layers.BatchNormalization())model.add(tf.keras.layers.Activation('relu'))model.add(tf.keras.layers.Dense(32))# model.add(tf.keras.layers.BatchNormalization())model.add(tf.keras.layers.Activation('relu'))model.add(tf.keras.layers.Dense(16))# model.add(tf.keras.layers.BatchNormalization())model.add(tf.keras.layers.Activation('relu'))model.add(tf.keras.layers.Dense(1, activation='sigmoid'))return if __name__ == '__main__':model_file = './my_model.h5'if (os.path.isfile(model_file)):print('model file detected. Loading.')model = tf.keras.models.load_model(model_file)else:print('No model file detected. Starting from scratch.')model = build_model()pile(loss='binary_crossentropy', optimizer="adam", metrics=['accuracy'])model.fit(X_train, y_train, batch_size=100, epochs=1, validation_data=(X_test, y_test))
模型输出为 my_model.h5 , 准确率百分之90%
三. 导出tensorflow-serving 模型, 运行在线预测服务
def save_model_for_production(model, version, path='prod_models'):tf.keras.backend.set_learning_phase(1)if not os.path.exists(path):os.mkdir(path)export_path = os.path.join(pat.as_bytes(path),pat.as_bytes(version))builder = tf.saved_model.builder.SavedModelBuilder(export_path)model_input = tf.saved_model.utils.build_tensor_info(model.input)model_output = tf.saved_model.utils.build_tensor_info(model.output)prediction_signature = (tf.saved_model.signature_def_utils.build_signature_def(inputs={'inputs': model_input},outputs={'output': model_output},method_name=tf.saved_model.signature_constants.PREDICT_METHOD_NAME))with tf.keras.backend.get_session() as sess:builder.add_meta_graph_and_variables(sess=sess, tags=[tf.saved_model.tag_constants.SERVING],signature_def_map={'predict':prediction_signature,})builder.save()
导出为 tensorflow serving 模型
export_path = "tf-model"save_model_for_production(model, "7", export_path)
运行在线预测服务(tensorflow 官方方法)
/serving/bazel-bin/tensorflow_serving/model_servers/tensorflow_model_server --port=9000 --model_name=username --model_base_path=/data/model/tf-model
四. client通过grpc 调用预测服务
#!/usr/bin/env python # encoding: utf-8 """ @version: v1.0 @author: zwqjoy @contact: zwqjoy@ @site: /zwqjoy @file: client@time: /6/29 15:02 """from grpc.beta import implementationsimport tensorflow as tffrom tensorflow_serving.apis import predict_pb2from tensorflow_serving.apis import prediction_service_pb2import numpy as nptf.app.flags.DEFINE_string('server', '172.xxx.xxx.xxx:9000','PredictionService host:port')tf.app.flags.DEFINE_string('username', 'demo_user','传入一个username')FLAGS = tf.app.flags.FLAGSdef nametovec(username):username = username.astype(str)length = len(username)... ...return np.array([length, ...])def main(_):host, port = FLAGS.server.split(':')channel = implementations.insecure_channel(host, int(port))stub = prediction_service_pb2.beta_create_PredictionService_stub(channel)# Send request# See prediction_service.proto for gRPC request/response details.data = nametovec(FLAGS.username)data = data.astype(np.float32)request = predict_pb2.PredictRequest()request.model_spec.name = 'username' # 这个name跟tensorflow_model_server --model_name="username" 对应request.model_spec.signature_name = 'predict' # 这个signature_name 跟signature_def_map 对应request.inputs['inputs'].CopyFrom(tf.contrib.util.make_tensor_proto(data, shape=(1, 18)))result = stub.Predict(request, 10.0) # 10 secs timeoutprint(result)if __name__ == '__main__':tf.app.run()