def sepTests():
    gestures = ["circle", "l", "m", "round", "x", "z"]
    count = 0
    for gesture in gestures:
        data = dataflow.dataflow("data", gesture)
        test = data.get_test_xyz()
        for i in range(test.shape[1]):
            count = count + 1
            Ef = open("data\\tests\\" + str(count) + ".csv", "w")
            Ewriter = csv.writer(Ef, delimiter = ',', quotechar = '', quoting = csv.QUOTE_NONE, dialect = csv.unix_dialect)
            Ewriter.writerows(test[:,i,:])
            Ef.close()
            
        print(test.shape)
예제 #2
0
def main():
    print(math.log(2.14727368714e-64))
    return
    path = "data" + os.sep + "unitytrain"
    name = "circle-r-cw"
    dtf = dataflow(path, name)
    
    (E, P, Pi, cent, thr) = dtf.load_model()
    testFile = "2014-3-31--16-50-44.csv"
    alltests = get_xyz_data(path + os.sep + "circle-r-cw" + os.sep + testFile)
    joint = "Hand_R"
    
    test = alltests[joint]
    clusters = numpy.empty(len(test))
    #f = open ("output.txt", "wt")
    for i in range(len(test)):
        clusters[i] = get_point_cluster(test[i,:], cent)
       # f.write(str(clusters[i]) + "\n")
        
    
    score = pr_hmm(clusters, P, E, Pi)
    print(score)
예제 #3
0
def main ():

    data = dataflow.dataflow("data", gesture)
    training = data.get_train_xyz()
    testing = data.get_test_xyz()
    '''
    ****************************************************
    *  Initializing
    ****************************************************
    '''
    TRTS = classifier.classifier()
    
    centroids = TRTS.get_point_centroids(training, N, D)
    cent2 = TRTS.get_point_centroids(testing, N, D)
    ATrainBinned = TRTS.get_point_clusters(training, centroids, D)
    ATestBinned = TRTS.get_point_clusters(testing, centroids, D)
    ATestBinned2 = TRTS.get_point_clusters(testing, cent2, D)


    '''
    ****************************************************
    *  Training
    ****************************************************
    '''
    
    # Set priors
    pP = TRTS.prior_transition_matrix(M, LR)
    
    # Train the model:
    b = [x for x in range(N)]
    cyc = 5
    E, P, Pi, LL = TRTS.dhmm_numeric(ATrainBinned, pP, b, M, cyc, .00001)
    
    '''
    ****************************************************
    *  Testing
    ****************************************************
    '''
    
    sumLik = 0
    minLik = numpy.Infinity
    
    ATrainBinned = ATrainBinned[:,:,0]
    ATestBinned = ATestBinned[:,:,0]

    for j in range(len(ATrainBinned)):
        lik = TRTS.pr_hmm(ATrainBinned[j], P, E.transpose(), Pi)
        if lik < minLik:
            minLik = lik
        sumLik = sumLik + lik
    
    gestureRecThreshold = 2.0 * sumLik / len(ATrainBinned)
	
	#Uncomment the following to store the models.
   # data.store_model(E, P, Pi, centroids, gestureRecThreshold)
#     data.store_Binned(ATrainBinned[:,:,0],ATestBinned[:,:,0])
    
    
    print('\n********************************************************************')
    print('Testing %d sequences for a log likelihood greater than %.4f' % (len(ATestBinned), gestureRecThreshold))
    print('********************************************************************\n')
    
    recs = 0
    tLL = numpy.zeros(shape=(len(ATestBinned)))
    for j in range(len(ATestBinned)):

        tLL[j] = TRTS.pr_hmm(ATestBinned[j], P, E.transpose(), Pi)
        if tLL[j] > gestureRecThreshold:
            recs = recs + 1
            print("Log Likelihood: %.3f > %.3f (threshold) -- FOUND %s Gesture" % (tLL[j], gestureRecThreshold, gesture))
        else:
            print("Log Likelihood: %.3f < %.3f (threshold) -- NO %s Gesture" % (tLL[j], gestureRecThreshold, gesture))
        
    print('Recognition success rate: %.2f percent\n' % (100 * recs / len(ATestBinned)))