Classifying bodily exercise from smartphone information

Introduction

On this put up we’ll describe how one can use smartphone accelerometer and gyroscope information to foretell the bodily actions of the people carrying the telephones. The information used on this put up comes from the Smartphone-Primarily based Recognition of Human Actions and Postural Transitions Information Set distributed by the College of California, Irvine. Thirty people have been tasked with performing numerous fundamental actions with an hooked up smartphone recording motion utilizing an accelerometer and gyroscope.

Earlier than we start, let’s load the varied libraries that we’ll use within the evaluation:

library(keras)     # Neural Networks
library(tidyverse) # Information cleansing / Visualization
library(knitr)     # Desk printing
library(rmarkdown) # Misc. output utilities 
library(ggridges)  # Visualization

Actions dataset

The information used on this put up come from the Smartphone-Primarily based Recognition of Human Actions and Postural Transitions Information Set(Reyes-Ortiz et al. 2016) distributed by the College of California, Irvine.

When downloaded from the hyperlink above, the information accommodates two completely different ‘elements.’ One which has been pre-processed utilizing numerous characteristic extraction methods corresponding to fast-fourier rework, and one other RawData part that merely provides the uncooked X,Y,Z instructions of an accelerometer and gyroscope. None of the usual noise filtering or characteristic extraction utilized in accelerometer information has been utilized. That is the information set we’ll use.

The motivation for working with the uncooked information on this put up is to assist the transition of the code/ideas to time collection information in much less well-characterized domains. Whereas a extra correct mannequin may very well be made by using the filtered/cleaned information supplied, the filtering and transformation can range vastly from activity to activity; requiring numerous guide effort and area data. One of many lovely issues about deep studying is the characteristic extraction is discovered from the information, not outdoors data.

Exercise labels

The information has integer encodings for the actions which, whereas not necessary to the mannequin itself, are useful to be used to see. Let’s load them first.

activityLabels <- learn.desk("information/activity_labels.txt", 
                             col.names = c("quantity", "label")) 

activityLabels %>% kable(align = c("c", "l"))

Subsequent, we load within the labels key for the RawData. This file is an inventory of all the observations, or particular person exercise recordings, contained within the information set. The important thing for the columns is taken from the information README.txt.

Column 1: experiment quantity ID, 
Column 2: consumer quantity ID, 
Column 3: exercise quantity ID 
Column 4: Label begin level 
Column 5: Label finish level 

The beginning and finish factors are in variety of sign log samples (recorded at 50hz).

Let’s check out the primary 50 rows:

labels <- learn.desk(
  "information/RawData/labels.txt",
  col.names = c("experiment", "userId", "exercise", "startPos", "endPos")
)

labels %>% 
  head(50) %>% 
  paged_table()

File names

Subsequent, let’s have a look at the precise information of the consumer information supplied to us in RawData/

dataFiles <- record.information("information/RawData")
dataFiles %>% head()

[1] "acc_exp01_user01.txt" "acc_exp02_user01.txt"
[3] "acc_exp03_user02.txt" "acc_exp04_user02.txt"
[5] "acc_exp05_user03.txt" "acc_exp06_user03.txt"

There’s a three-part file naming scheme. The primary half is the kind of information the file accommodates: both acc for accelerometer or gyro for gyroscope. Subsequent is the experiment quantity, and final is the consumer Id for the recording. Let’s load these right into a dataframe for ease of use later.

fileInfo <- data_frame(
  filePath = dataFiles
) %>%
  filter(filePath != "labels.txt") %>% 
  separate(filePath, sep = '_', 
           into = c("sort", "experiment", "userId"), 
           take away = FALSE) %>% 
  mutate(
    experiment = str_remove(experiment, "exp"),
    userId = str_remove_all(userId, "consumer|.txt")
  ) %>% 
  unfold(sort, filePath)

fileInfo %>% head() %>% kable()

Studying and gathering information

Earlier than we are able to do something with the information supplied we have to get it right into a model-friendly format. This implies we wish to have an inventory of observations, their class (or exercise label), and the information akin to the recording.

To acquire this we’ll scan by means of every of the recording information current in dataFiles, search for what observations are contained within the recording, extract these recordings and return all the things to a simple to mannequin with dataframe.

# Learn contents of single file to a dataframe with accelerometer and gyro information.
readInData <- operate(experiment, userId){
  genFilePath = operate(sort) {
    paste0("information/RawData/", sort, "_exp",experiment, "_user", userId, ".txt")
  }  
  
  bind_cols(
    learn.desk(genFilePath("acc"), col.names = c("a_x", "a_y", "a_z")),
    learn.desk(genFilePath("gyro"), col.names = c("g_x", "g_y", "g_z"))
  )
}

# Perform to learn a given file and get the observations contained alongside
# with their lessons.

loadFileData <- operate(curExperiment, curUserId) {
  
  # load sensor information from file into dataframe
  allData <- readInData(curExperiment, curUserId)

  extractObservation <- operate(startPos, endPos){
    allData[startPos:endPos,]
  }
  
  # get remark places on this file from labels dataframe
  dataLabels <- labels %>% 
    filter(userId == as.integer(curUserId), 
           experiment == as.integer(curExperiment))
  

  # extract observations as dataframes and save as a column in dataframe.
  dataLabels %>% 
    mutate(
      information = map2(startPos, endPos, extractObservation)
    ) %>% 
    choose(-startPos, -endPos)
}

# scan by means of all experiment and userId combos and collect information right into a dataframe. 
allObservations <- map2_df(fileInfo$experiment, fileInfo$userId, loadFileData) %>% 
  right_join(activityLabels, by = c("exercise" = "quantity")) %>% 
  rename(activityName = label)

# cache work. 
write_rds(allObservations, "allObservations.rds")
allObservations %>% dim()

Exploring the information

Now that we’ve got all the information loaded together with the experiment, userId, and exercise labels, we are able to discover the information set.

Size of recordings

Let’s first have a look at the size of the recordings by exercise.

allObservations %>% 
  mutate(recording_length = map_int(information,nrow)) %>% 
  ggplot(aes(x = recording_length, y = activityName)) +
  geom_density_ridges(alpha = 0.8)

The very fact there may be such a distinction in size of recording between the completely different exercise sorts requires us to be a bit cautious with how we proceed. If we prepare the mannequin on each class directly we’re going to need to pad all of the observations to the size of the longest, which would go away a big majority of the observations with an enormous proportion of their information being simply padding-zeros. Due to this, we’ll match our mannequin to only the biggest ‘group’ of observations size actions, these embody STAND_TO_SIT, STAND_TO_LIE, SIT_TO_STAND, SIT_TO_LIE, LIE_TO_STAND, and LIE_TO_SIT.

An fascinating future path could be trying to make use of one other structure corresponding to an RNN that may deal with variable size inputs and coaching it on all the information. Nonetheless, you’ll run the danger of the mannequin studying merely that if the remark is lengthy it’s most definitely one of many 4 longest lessons which might not generalize to a situation the place you have been operating this mannequin on a real-time-stream of information.

Filtering actions

Primarily based on our work from above, let’s subset the information to only be of the actions of curiosity.

desiredActivities <- c(
  "STAND_TO_SIT", "SIT_TO_STAND", "SIT_TO_LIE", 
  "LIE_TO_SIT", "STAND_TO_LIE", "LIE_TO_STAND"  
)

filteredObservations <- allObservations %>% 
  filter(activityName %in% desiredActivities) %>% 
  mutate(observationId = 1:n())

filteredObservations %>% paged_table()

So after our aggressive pruning of the information we can have a decent quantity of information left upon which our mannequin can be taught.

Coaching/testing cut up

Earlier than we go any additional into exploring the information for our mannequin, in an try to be as truthful as potential with our efficiency measures, we have to cut up the information right into a prepare and check set. Since every consumer carried out all actions simply as soon as (except one who solely did 10 of the 12 actions) by splitting on userId we’ll be certain that our mannequin sees new individuals completely after we check it.

# get all customers
userIds <- allObservations$userId %>% distinctive()

# randomly select 24 (80% of 30 people) for coaching
set.seed(42) # seed for reproducibility
trainIds <- pattern(userIds, dimension = 24)

# set the remainder of the customers to the testing set
testIds <- setdiff(userIds,trainIds)

# filter information. 
trainData <- filteredObservations %>% 
  filter(userId %in% trainIds)

testData <- filteredObservations %>% 
  filter(userId %in% testIds)

Visualizing actions

Now that we’ve got trimmed our information by eradicating actions and splitting off a check set, we are able to really visualize the information for every class to see if there’s any instantly discernible form that our mannequin might be able to decide up on.

First let’s unpack our information from its dataframe of one-row-per-observation to a tidy model of all of the observations.

unpackedObs <- 1:nrow(trainData) %>% 
  map_df(operate(rowNum){
    dataRow <- trainData[rowNum, ]
    dataRow$information[[1]] %>% 
      mutate(
        activityName = dataRow$activityName, 
        observationId = dataRow$observationId,
        time = 1:n() )
  }) %>% 
  collect(studying, worth, -time, -activityName, -observationId) %>% 
  separate(studying, into = c("sort", "path"), sep = "_") %>% 
  mutate(sort = ifelse(sort == "a", "acceleration", "gyro"))

Now we’ve got an unpacked set of our observations, let’s visualize them!

unpackedObs %>% 
  ggplot(aes(x = time, y = worth, shade = path)) +
  geom_line(alpha = 0.2) +
  geom_smooth(se = FALSE, alpha = 0.7, dimension = 0.5) +
  facet_grid(sort ~ activityName, scales = "free_y") +
  theme_minimal() +
  theme( axis.textual content.x = element_blank() )

So at the very least within the accelerometer information patterns undoubtedly emerge. One would think about that the mannequin might have hassle with the variations between LIE_TO_SIT and LIE_TO_STAND as they’ve an identical profile on common. The identical goes for SIT_TO_STAND and STAND_TO_SIT.

Preprocessing

Earlier than we are able to prepare the neural community, we have to take a few steps to preprocess the information.

Padding observations

First we’ll determine what size to pad (and truncate) our sequences to by discovering what the 98th percentile size is. By not utilizing the very longest remark size this can assist us keep away from extra-long outlier recordings messing up the padding.

padSize <- trainData$information %>% 
  map_int(nrow) %>% 
  quantile(p = 0.98) %>% 
  ceiling()
padSize

98% 
334 

Now we merely have to convert our record of observations to matrices, then use the tremendous useful pad_sequences() operate in Keras to pad all observations and switch them right into a 3D tensor for us.

convertToTensor <- . %>% 
  map(as.matrix) %>% 
  pad_sequences(maxlen = padSize)

trainObs <- trainData$information %>% convertToTensor()
testObs <- testData$information %>% convertToTensor()
  
dim(trainObs)

[1] 286 334   6

Great, we now have our information in a pleasant neural-network-friendly format of a 3D tensor with dimensions (, , ).

One-hot encoding

There’s one last item we have to do earlier than we are able to prepare our mannequin, and that’s flip our remark lessons from integers into one-hot, or dummy encoded, vectors. Fortunately, once more Keras has equipped us with a really useful operate to just do this.

oneHotClasses <- . %>% 
  {. - 7} %>%        # convey integers right down to 0-6 from 7-12
  to_categorical() # One-hot encode

trainY <- trainData$exercise %>% oneHotClasses()
testY <- testData$exercise %>% oneHotClasses()

Modeling

Structure

Since we’ve got temporally dense time-series information we’ll make use of 1D convolutional layers. With temporally-dense information, an RNN has to be taught very lengthy dependencies as a way to decide up on patterns, CNNs can merely stack a number of convolutional layers to construct sample representations of considerable size. Since we’re additionally merely searching for a single classification of exercise for every remark, we are able to simply use pooling to ‘summarize’ the CNNs view of the information right into a dense layer.

Along with stacking two layer_conv_1d() layers, we’ll use batch norm and dropout (the spatial variant(Tompson et al. 2014) on the convolutional layers and customary on the dense) to regularize the community.

input_shape <- dim(trainObs)[-1]
num_classes <- dim(trainY)[2]

filters <- 24     # variety of convolutional filters to be taught
kernel_size <- 8  # what number of time-steps every conv layer sees.
dense_size <- 48  # dimension of our penultimate dense layer. 

# Initialize mannequin
mannequin <- keras_model_sequential()
mannequin %>% 
  layer_conv_1d(
    filters = filters,
    kernel_size = kernel_size, 
    input_shape = input_shape,
    padding = "legitimate", 
    activation = "relu"
  ) %>%
  layer_batch_normalization() %>%
  layer_spatial_dropout_1d(0.15) %>% 
  layer_conv_1d(
    filters = filters/2,
    kernel_size = kernel_size,
    activation = "relu",
  ) %>%
  # Apply common pooling:
  layer_global_average_pooling_1d() %>% 
  layer_batch_normalization() %>%
  layer_dropout(0.2) %>% 
  layer_dense(
    dense_size,
    activation = "relu"
  ) %>% 
  layer_batch_normalization() %>%
  layer_dropout(0.25) %>% 
  layer_dense(
    num_classes, 
    activation = "softmax",
    identify = "dense_output"
  ) 

abstract(mannequin)

______________________________________________________________________
Layer (sort)                   Output Form                Param #    
======================================================================
conv1d_1 (Conv1D)              (None, 327, 24)             1176       
______________________________________________________________________
batch_normalization_1 (BatchNo (None, 327, 24)             96         
______________________________________________________________________
spatial_dropout1d_1 (SpatialDr (None, 327, 24)             0          
______________________________________________________________________
conv1d_2 (Conv1D)              (None, 320, 12)             2316       
______________________________________________________________________
global_average_pooling1d_1 (Gl (None, 12)                  0          
______________________________________________________________________
batch_normalization_2 (BatchNo (None, 12)                  48         
______________________________________________________________________
dropout_1 (Dropout)            (None, 12)                  0          
______________________________________________________________________
dense_1 (Dense)                (None, 48)                  624        
______________________________________________________________________
batch_normalization_3 (BatchNo (None, 48)                  192        
______________________________________________________________________
dropout_2 (Dropout)            (None, 48)                  0          
______________________________________________________________________
dense_output (Dense)           (None, 6)                   294        
======================================================================
Complete params: 4,746
Trainable params: 4,578
Non-trainable params: 168
______________________________________________________________________

Coaching

Now we are able to prepare the mannequin utilizing our check and coaching information. Notice that we use callback_model_checkpoint() to make sure that we save solely the most effective variation of the mannequin (fascinating since in some unspecified time in the future in coaching the mannequin might start to overfit or in any other case cease enhancing).

# Compile mannequin
mannequin %>% compile(
  loss = "categorical_crossentropy",
  optimizer = "rmsprop",
  metrics = "accuracy"
)

trainHistory <- mannequin %>%
  match(
    x = trainObs, y = trainY,
    epochs = 350,
    validation_data = record(testObs, testY),
    callbacks = record(
      callback_model_checkpoint("best_model.h5", 
                                save_best_only = TRUE)
    )
  )

The mannequin is studying one thing! We get a decent 94.4% accuracy on the validation information, not unhealthy with six potential lessons to select from. Let’s look into the validation efficiency just a little deeper to see the place the mannequin is messing up.

Analysis

Now that we’ve got a educated mannequin let’s examine the errors that it made on our testing information. We will load the most effective mannequin from coaching based mostly upon validation accuracy after which have a look at every remark, what the mannequin predicted, how excessive a likelihood it assigned, and the true exercise label.

# dataframe to get labels onto one-hot encoded prediction columns
oneHotToLabel <- activityLabels %>% 
  mutate(quantity = quantity - 7) %>% 
  filter(quantity >= 0) %>% 
  mutate(class = paste0("V",quantity + 1)) %>% 
  choose(-number)

# Load our greatest mannequin checkpoint
bestModel <- load_model_hdf5("best_model.h5")

tidyPredictionProbs <- bestModel %>% 
  predict(testObs) %>% 
  as_data_frame() %>% 
  mutate(obs = 1:n()) %>% 
  collect(class, prob, -obs) %>% 
  right_join(oneHotToLabel, by = "class")

predictionPerformance <- tidyPredictionProbs %>% 
  group_by(obs) %>% 
  summarise(
    highestProb = max(prob),
    predicted = label[prob == highestProb]
  ) %>% 
  mutate(
    reality = testData$activityName,
    right = reality == predicted
  ) 

predictionPerformance %>% paged_table()

First, let’s have a look at how ‘assured’ the mannequin was by if the prediction was right or not.

predictionPerformance %>% 
  mutate(consequence = ifelse(right, 'Appropriate', 'Incorrect')) %>% 
  ggplot(aes(highestProb)) +
  geom_histogram(binwidth = 0.01) +
  geom_rug(alpha = 0.5) +
  facet_grid(consequence~.) +
  ggtitle("Chances related to prediction by correctness")

Reassuringly it appears the mannequin was, on common, much less assured about its classifications for the inaccurate outcomes than the right ones. (Though, the pattern dimension is just too small to say something definitively.)

Let’s see what actions the mannequin had the toughest time with utilizing a confusion matrix.

predictionPerformance %>% 
  group_by(reality, predicted) %>% 
  summarise(depend = n()) %>% 
  mutate(good = reality == predicted) %>% 
  ggplot(aes(x = reality,  y = predicted)) +
  geom_point(aes(dimension = depend, shade = good)) +
  geom_text(aes(label = depend), 
            hjust = 0, vjust = 0, 
            nudge_x = 0.1, nudge_y = 0.1) + 
  guides(shade = FALSE, dimension = FALSE) +
  theme_minimal()

We see that, because the preliminary visualization advised, the mannequin had a little bit of hassle with distinguishing between LIE_TO_SIT and LIE_TO_STAND lessons, together with the SIT_TO_LIE and STAND_TO_LIE, which even have comparable visible profiles.

Future instructions

The obvious future path to take this evaluation could be to aim to make the mannequin extra basic by working with extra of the equipped exercise sorts. One other fascinating path could be to not separate the recordings into distinct ‘observations’ however as an alternative maintain them as one streaming set of information, very similar to an actual world deployment of a mannequin would work, and see how effectively a mannequin might classify streaming information and detect adjustments in exercise.