Пример #1
0
def get_features(labeled=False):
    mkdir.mkdir("../out/{0}".format(WHICH_EXP))

    if labeled:
        good_tracks, data_panel, feat_space, feature_names, labels_panel = feature_setup(labeled)
    else:
        good_tracks, data_panel, feat_space, feature_names = feature_setup()

    features_dict = {}
    for i in good_tracks:
        pos = np.array([data_panel[i].dropna(axis=0)["x"], data_panel[i].dropna(axis=0)["y"]])
        pos = np.transpose(pos)
        features_dict[i] = pd.DataFrame(feat_space.get_features(pos), columns=feature_names)

    features_panel = pd.Panel(features_dict)
    cPickle.dump(features_panel, open("../out/{0}/testingkthx.pk".format(WHICH_EXP), "w"))

    # make feat_array and label_arrays so that the following loop can
    # use vstack and hstack
    # we need the arrays to make histograms and such
    feat_array = np.vstack(
        [features_panel[good_tracks[0]].dropna(axis=0)[:], features_panel[good_tracks[1]].dropna(axis=0)[:]]
    )
    if labeled:
        label_array = np.hstack(
            [
                labels_panel[good_tracks[0]].dropna(axis=0)["labels"],
                labels_panel[good_tracks[1]].dropna(axis=0)["labels"],
            ]
        )

    for i in good_tracks[2:]:
        feat_array = np.vstack([feat_array, features_panel[i].dropna(axis=0)[:]])
        if labeled:
            label_array = np.hstack([label_array, labels_panel[i].dropna(axis=0)["labels"]])
    if labeled:
        label_array = np.transpose(label_array)

    pairs = []
    for track in good_tracks:
        for i in range(len(data_panel[track].dropna(axis=0).index)):
            pairs.append((track, i))

    multi = pd.MultiIndex.from_tuples(pairs, names=["track", "frame"])

    X = pd.DataFrame(feat_array, index=multi, columns=feature_names)
    cPickle.dump(X, open("../out/{0}/X.pk".format(WHICH_EXP), "w"))

    if labeled:
        Y = pd.DataFrame(label_array, index=multi, columns=["labels"])
        cPickle.dump(Y, open("../out/{0}/Y_supervised.pk".format(WHICH_EXP), "w"))
Пример #2
0
def plot(labeled=False):
    X = load_data.X()

    if labeled:
        Y_supervised = load_data.Y_supervised()
        good_tracks, data_panel,feat_space, feature_names, labels_panel = feature_setup(labeled)
        plot_hist(X,Y_supervised,'Human Labeled','supervised_hist')
    else:
        good_tracks, data_panel,feat_space, feature_names = feature_setup(labeled)
    
    Y_hmm = load_data.Y_hmm()
    plot_hist(X,Y_hmm,'HMM','hmm_hist')

    Y_svm = load_data.Y_svm()
    plot_hist(X,Y_svm,'SVM','svm_hist')