聲明：版權所有，轉載請聯系作者并注明出處??http://blog.csdn.net/u013719780?viewmode=contents

Autoencoders and Neural Network for Place recognition with WiFi fingerprints

本文來源于Micha? Nowicki?and Jan Wietrzykowski?論文的讀書筆記

論文原文：https://arxiv.org/pdf/1611.02049v1.pdf

現實世界的很多場景需要知道用戶位置以便為他們提供某些服務。因此,自動用戶定位一直是近年來的研究熱點。自動用戶定位包括估算用戶的位置(緯度、經度和海拔)。由于有包括GPS傳感器等連接在移動設備上，解決室外定位問題比較容易。然而,室內定位還面臨這多困難,仍然是一個懸而未決的問題，主要是由于在室內環境中GPS信號有損失。

本文使用數據集UJIIndoorLoc進行實驗。數據集的詳細信息如下：

Attribute Information:

Attribute 001 (WAP001): Intensity value for WAP001. Negative integer values from -104 to 0 and +100. Positive value 100 used if WAP001 was not detected.
....
Attribute 520 (WAP520): Intensity value for WAP520. Negative integer values from -104 to 0 and +100. Positive Vvalue 100 used if WAP520 was not detected.
Attribute 521 (Longitude): Longitude. Negative real values from -7695.9387549299299000 to -7299.786516730871000
Attribute 522 (Latitude): Latitude. Positive real values from 4864745.7450159714 to 4865017.3646842018.
Attribute 523 (Floor): Altitude in floors inside the building. Integer values from 0 to 4.
Attribute 524 (BuildingID): ID to identify the building. Measures were taken in three different buildings. Categorical integer values from 0 to 2.
Attribute 525 (SpaceID): Internal ID number to identify the Space (office, corridor, classroom) where the capture was taken. Categorical integer values.
Attribute 526 (RelativePosition): Relative position with respect to the Space (1 - Inside, 2 - Outside in Front of the door). Categorical integer values.
Attribute 527 (UserID): User identifier (see below). Categorical integer values.
Attribute 528 (PhoneID): Android device identifier (see below). Categorical integer values.
Attribute 529 (Timestamp): UNIX Time when the capture was taken. Integer value.

---

UserID Anonymized user Height (cm)?

0 USER0000 (Validation User) N/A
1 USER0001 170
2 USER0002 176
3 USER0003 172
4 USER0004 174
5 USER0005 184
6 USER0006 180
7 USER0007 160
8 USER0008 176
9 USER0009 177
10 USER0010 186
11 USER0011 176
12 USER0012 158
13 USER0013 174
14 USER0014 173
15 USER0015 174
16 USER0016 171
17 USER0017 166

18 USER0018 162?

---

PhoneID Android Device Android Ver. UserID?

0 Celkon A27 4.0.4(6577) 0
1 GT-I8160 2.3.6 8
2 GT-I8160 4.1.2 0
3 GT-I9100 4.0.4 5
4 GT-I9300 4.1.2 0
5 GT-I9505 4.2.2 0
6 GT-S5360 2.3.6 7
7 GT-S6500 2.3.6 14
8 Galaxy Nexus 4.2.2 10
9 Galaxy Nexus 4.3 0
10 HTC Desire HD 2.3.5 18
11 HTC One 4.1.2 15
12 HTC One 4.2.2 0
13 HTC Wildfire S 2.3.5 0,11
14 LT22i 4.0.4 0,1,9,16
15 LT22i 4.1.2 0
16 LT26i 4.0.4 3
17 M1005D 4.0.4 13
18 MT11i 2.3.4 4
19 Nexus 4 4.2.2 6
20 Nexus 4 4.3 0
21 Nexus S 4.1.2 0
22 Orange Monte Carlo 2.3.5 17
23 Transformer TF101 4.0.3 2
24 bq Curie 4.1.1 12

In?[1]: import pandas as pd import numpy as np import tensorflow as tf from sklearn.preprocessing import scale In?[2]: dataset = pd.read_csv("trainingData.csv",header = 0) features = scale(np.asarray(dataset.ix[:,0:520])) labels = np.asarray(dataset["BUILDINGID"].map(str) + dataset["FLOOR"].map(str)) labels = np.asarray(pd.get_dummies(labels)) /Applications/anaconda/lib/python2.7/site-packages/sklearn/utils/validation.py:420: DataConversionWarning: Data with input dtype int64 was converted to float64 by the scale function.warnings.warn(msg, DataConversionWarning)

Dividing UJIndoorLoc training data set into training and validation set

In?[3]: train_val_split = np.random.rand(len(features)) < 0.70 train_x = features[train_val_split] train_y = labels[train_val_split] val_x = features[~train_val_split] val_y = labels[~train_val_split]

Using UJIndoorLoc validation data set as testing set

In?[4]: test_dataset = pd.read_csv("validationData.csv",header = 0) test_features = scale(np.asarray(test_dataset.ix[:,0:520])) test_labels = np.asarray(test_dataset["BUILDINGID"].map(str) + test_dataset["FLOOR"].map(str)) test_labels = np.asarray(pd.get_dummies(test_labels)) /Applications/anaconda/lib/python2.7/site-packages/sklearn/utils/validation.py:420: DataConversionWarning: Data with input dtype int64 was converted to float64 by the scale function.warnings.warn(msg, DataConversionWarning) In?[5]: def weight_variable(shape):initial = tf.truncated_normal(shape, stddev = 0.1)return tf.Variable(initial)def bias_variable(shape):initial = tf.constant(0.0, shape = shape)return tf.Variable(initial) In?[6]: n_input = 520 n_hidden_1 = 256 n_hidden_2 = 128 n_hidden_3 = 64 n_classes = labels.shape[1]learning_rate = 0.01 training_epochs = 20 batch_size = 10total_batches = dataset.shape[0] // batch_size In?[7]: X = tf.placeholder(tf.float32, shape=[None,n_input]) Y = tf.placeholder(tf.float32,[None,n_classes])# --------------------- Encoder Variables --------------- #e_weights_h1 = weight_variable([n_input, n_hidden_1]) e_biases_h1 = bias_variable([n_hidden_1])e_weights_h2 = weight_variable([n_hidden_1, n_hidden_2]) e_biases_h2 = bias_variable([n_hidden_2])e_weights_h3 = weight_variable([n_hidden_2, n_hidden_3]) e_biases_h3 = bias_variable([n_hidden_3])# --------------------- Decoder Variables --------------- #d_weights_h1 = weight_variable([n_hidden_3, n_hidden_2]) d_biases_h1 = bias_variable([n_hidden_2])d_weights_h2 = weight_variable([n_hidden_2, n_hidden_1]) d_biases_h2 = bias_variable([n_hidden_1])d_weights_h3 = weight_variable([n_hidden_1, n_input]) d_biases_h3 = bias_variable([n_input])# --------------------- DNN Variables ------------------ #dnn_weights_h1 = weight_variable([n_hidden_3, n_hidden_2]) dnn_biases_h1 = bias_variable([n_hidden_2])dnn_weights_h2 = weight_variable([n_hidden_2, n_hidden_2]) dnn_biases_h2 = bias_variable([n_hidden_2])dnn_weights_out = weight_variable([n_hidden_2, n_classes]) dnn_biases_out = bias_variable([n_classes]) In?[8]: def encode(x):l1 = tf.nn.tanh(tf.add(tf.matmul(x,e_weights_h1),e_biases_h1))l2 = tf.nn.tanh(tf.add(tf.matmul(l1,e_weights_h2),e_biases_h2))l3 = tf.nn.tanh(tf.add(tf.matmul(l2,e_weights_h3),e_biases_h3))return l3def decode(x):l1 = tf.nn.tanh(tf.add(tf.matmul(x,d_weights_h1),d_biases_h1))l2 = tf.nn.tanh(tf.add(tf.matmul(l1,d_weights_h2),d_biases_h2))l3 = tf.nn.tanh(tf.add(tf.matmul(l2,d_weights_h3),d_biases_h3))return l3def dnn(x):l1 = tf.nn.tanh(tf.add(tf.matmul(x,dnn_weights_h1),dnn_biases_h1))l2 = tf.nn.tanh(tf.add(tf.matmul(l1,dnn_weights_h2),dnn_biases_h2))out = tf.nn.softmax(tf.add(tf.matmul(l2,dnn_weights_out),dnn_biases_out))return out In?[9]: encoded = encode(X) decoded = decode(encoded) y_ = dnn(encoded) In?[10]: us_cost_function = tf.reduce_mean(tf.pow(X - decoded, 2)) s_cost_function = -tf.reduce_sum(Y * tf.log(y_)) us_optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(us_cost_function) s_optimizer = tf.train.GradientDescentOptimizer(learning_rate).minimize(s_cost_function) In?[11]: correct_prediction = tf.equal(tf.argmax(y_,1), tf.argmax(Y,1)) accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

模型結構

圖片取自論文原文:?https://arxiv.org/pdf/1611.02049v1.pdf

In?[12]: with tf.Session() as session:tf.initialize_all_variables().run()# ------------ 1. Training Autoencoders - Unsupervised Learning ----------- #for epoch in range(training_epochs):epoch_costs = np.empty(0)for b in range(total_batches):offset = (b * batch_size) % (features.shape[0] - batch_size)batch_x = features[offset:(offset + batch_size), :]_, c = session.run([us_optimizer, us_cost_function],feed_dict={X: batch_x})epoch_costs = np.append(epoch_costs,c)print "Epoch: ",epoch," Loss: ",np.mean(epoch_costs)print "Unsupervised pre-training finished..."# ---------------- 2. Training NN - Supervised Learning ------------------ #for epoch in range(training_epochs):epoch_costs = np.empty(0)for b in range(total_batches):offset = (b * batch_size) % (train_x.shape[0] - batch_size)batch_x = train_x[offset:(offset + batch_size), :]batch_y = train_y[offset:(offset + batch_size), :]_, c = session.run([s_optimizer, s_cost_function],feed_dict={X: batch_x, Y : batch_y})epoch_costs = np.append(epoch_costs,c)print "Epoch: ",epoch," Loss: ",np.mean(epoch_costs)," Training Accuracy: ", \session.run(accuracy, feed_dict={X: train_x, Y: train_y}), \"Validation Accuracy:", session.run(accuracy, feed_dict={X: val_x, Y: val_y})print "Supervised training finished..."print "\nTesting Accuracy:", session.run(accuracy, feed_dict={X: test_features, Y: test_labels}) Epoch: 0 Loss: 0.946417506465 Epoch: 1 Loss: 0.872724663348 Epoch: 2 Loss: 0.834939743301 Epoch: 3 Loss: 0.812426232725 Epoch: 4 Loss: 0.797482786783 Epoch: 5 Loss: 0.786710632076 Epoch: 6 Loss: 0.778407857031 Epoch: 7 Loss: 0.771705506587 Epoch: 8 Loss: 0.766137119669 Epoch: 9 Loss: 0.761425737017 Epoch: 10 Loss: 0.757388042674 Epoch: 11 Loss: 0.753892278756 Epoch: 12 Loss: 0.750838549624 Epoch: 13 Loss: 0.748148437356 Epoch: 14 Loss: 0.745759032577 Epoch: 15 Loss: 0.743619203037 Epoch: 16 Loss: 0.741687947782 Epoch: 17 Loss: 0.739931871977 Epoch: 18 Loss: 0.738323841833 Epoch: 19 Loss: 0.736841905176 Unsupervised pre-training finished... Epoch: 0 Loss: 4.27480528807 Training Accuracy: 0.507781 Validation Accuracy: 0.504976 Epoch: 1 Loss: 2.49841824394 Training Accuracy: 0.689392 Validation Accuracy: 0.675493 Epoch: 2 Loss: 1.63081804872 Training Accuracy: 0.772202 Validation Accuracy: 0.757801 Epoch: 3 Loss: 1.13079869366 Training Accuracy: 0.820174 Validation Accuracy: 0.80334 Epoch: 4 Loss: 0.957438500986 Training Accuracy: 0.845874 Validation Accuracy: 0.828808 Epoch: 5 Loss: 0.759610852076 Training Accuracy: 0.891919 Validation Accuracy: 0.871648 Epoch: 6 Loss: 0.668730932672 Training Accuracy: 0.834166 Validation Accuracy: 0.815483 Epoch: 7 Loss: 0.644024800692 Training Accuracy: 0.895131 Validation Accuracy: 0.869118 Epoch: 8 Loss: 0.586599285754 Training Accuracy: 0.926613 Validation Accuracy: 0.908416 Epoch: 9 Loss: 0.368939029375 Training Accuracy: 0.937607 Validation Accuracy: 0.914825 Epoch: 10 Loss: 0.383492833295 Training Accuracy: 0.932967 Validation Accuracy: 0.913307 Epoch: 11 Loss: 0.348123138894 Training Accuracy: 0.945174 Validation Accuracy: 0.918705 Epoch: 12 Loss: 0.397730887604 Training Accuracy: 0.948672 Validation Accuracy: 0.924776 Epoch: 13 Loss: 0.316421336578 Training Accuracy: 0.952741 Validation Accuracy: 0.924945 Epoch: 14 Loss: 0.29667654338 Training Accuracy: 0.95474 Validation Accuracy: 0.926632 Epoch: 15 Loss: 0.279483787604 Training Accuracy: 0.948101 Validation Accuracy: 0.920729 Epoch: 16 Loss: 0.331246199433 Training Accuracy: 0.971945 Validation Accuracy: 0.945859 Epoch: 17 Loss: 0.250471480337 Training Accuracy: 0.967875 Validation Accuracy: 0.943161 Epoch: 18 Loss: 0.221882153683 Training Accuracy: 0.970874 Validation Accuracy: 0.946365 Epoch: 19 Loss: 0.245788279902 Training Accuracy: 0.977013 Validation Accuracy: 0.950076 Supervised training finished...Testing Accuracy: 0.717372

總結

以上是生活随笔為你收集整理的机器学习实验（十）：基于WiFi fingerprints用自编码器(Autoencoders)和神经网络(Neural Network)进行定位_1(tensorflow版)的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。