pythonで加速度のデータから行動認識をやってみる① ref: http://qiita.com/LittleWat/items/f4156cffd714f2377bb0

dbからの読み込み

con = sqlite3.connect(dbBaseName + ".db") #dbBaseNameを指定してください
accDf = pd.read_sql_query("SELECT * from acc", con) #"acc"という名前に加速度データが記録してあります．
con.close()

featureExtraction.py

def calcFFTenergy(col):
    X = np.fft.fft(col)  # FFT
    squaredAmplitudeSpectrum = np.array([(c.real ** 2 + c.imag ** 2) for c in X])
    return squaredAmplitudeSpectrum.sum() / len(col)

def corrMat2series(df):
    """
    :param df: pd.DataFrame.corr()
    :return: pd.Series
    """
    import itertools as it
    combinationList = list(it.combinations(df.columns, 2))  # make 2 combinations
    name = ["corr_" + "".join(list(x)) for x in combinationList]
    corrArr = []
    for combination in combinationList:
        corrArr.append(df.loc[combination[0], combination[1]])
    corrSeries = pd.Series(corrArr, index=name)
    return corrSeries

windowSize = 128 #2のべき乗
slidingSize = int(windowSize/2 )

endIdx = smallDF.shape[0]
cnt = 0
outDF = pd.DataFrame()
while (cnt + windowSize < endIdx):
    windowDF = smallDF.iloc[cnt:cnt + windowSize, :]
    meanSeries = windowDF.mean().add_prefix("mean_")
    stdSeries = windowDF.std().add_prefix("std_")
    energySeries = windowDF.apply(calcFFTenergy).add_prefix("energy_")
    corrSeries = corrMat2series(windowDF.corr())
    allSeries = pd.concat([meanSeries, stdSeries, energySeries, corrSeries])
    allSeries.name = windowDF.index[0]
    outDF = outDF.append(allSeries)
    cnt += slidingSize

resample.py

smallDF.loc[:,"time"]  = smallDF.index
smallDF.index =smallDF.time.apply(lambda x: datetime.fromtimestamp(x/1000))  #データ型の変更
del smallDF["time"]
smallDF = smallDF.resample('33L', how='mean') #33msecでリサンプリング
smallDF = smallDF.interpolate() #若干の欠損が出るので，線形補間

SensingActivityの一部.java

//manager.registerListener(this, s, SensorManager.SENSOR_DELAY_GAME);
manager.registerListener(this, s, 20000); //20000μsec(=50Hz)

ライブラリのインポート

import pandas as pd
import numpy as np
import sqlite3 
from datetime import *