def plot_trajectory(ax, fontproperties, CC_dict, bootstrap_dict, crit_freq,
                    model_version):

    fonts = fontproperties
    size = fonts.get_size()
    markersize = 8

    if type(crit_freq) == tuple:
        band_name = str(crit_freq[0]) + '-' + str(crit_freq[1]) + ' Hz'
    else:
        band_name = '0.1 - ' + str(crit_freq) + ' Hz'

    ong_CCs = CC_dict['ongoing']
    ong_boots = bootstrap_dict['ongoing']
    trans_CCs = CC_dict['transient']
    trans_boots = bootstrap_dict['transient']
    ss_CCs = CC_dict['steady-state']
    ss_boots = bootstrap_dict['steady-state']
    #need left-right "jitter" to make overlapping dots distinguishable
    jitters = numpy.zeros(len(ong_CCs))
    for n, val in enumerate(jitters):
        jitters[n] += numpy.random.uniform(-0.05, 0.05)

    ### Get p-vals for stim modulation of population, draw on plot.

    ong_trans_p_val = wlcx(ong_CCs, trans_CCs)[1]
    trans_ss_p_val = wlcx(trans_CCs, ss_CCs)[1]
    ong_ss_p_val = wlcx(ong_CCs, ss_CCs)[1]
    ong_trans_stat = wlcx(ong_CCs, trans_CCs)[0]
    trans_ss_stat = wlcx(trans_CCs, ss_CCs)[0]
    ong_ss_stat = wlcx(ong_CCs, ss_CCs)[0]

    print '    <CC> ong = %s' % str(
        numpy.mean(ong_CCs)), '+/-', numpy.std(ong_CCs)
    print '      (P = %s; one-sided t-test)' % str(
        0.5 * ttest_1samp(ong_CCs, 0)[1])
    print '    <CC> trans = %s' % str(
        numpy.mean(trans_CCs)), '+/-', numpy.std(trans_CCs)
    print '      (P = %s; one-sided t-test)' % str(
        0.5 * ttest_1samp(trans_CCs, 0)[1])
    print '    <CC> ss = %s' % str(
        numpy.mean(ss_CCs)), '+/-', numpy.std(ss_CCs)
    print '      (P = %s; one-sided t-test)' % str(
        0.5 * ttest_1samp(ss_CCs, 0)[1])

    print '    ong --> trans P = %s' % str(
        ong_trans_p_val), '(stat = %s)' % str(ong_trans_stat)
    print '    trans --> ss P = %s' % str(trans_ss_p_val), '(stat = %s)' % str(
        trans_ss_stat)
    print '    ong --> ss P = %s' % str(ong_ss_p_val), '(stat = %s)' % str(
        ong_ss_stat)

    print '    corrected ong --> trans P = %s' % str(
        3 * ong_trans_p_val), '(stat = %s)' % str(ong_trans_stat)
    print '    corrected trans --> ss P = %s' % str(
        3 * trans_ss_p_val), '(stat = %s)' % str(trans_ss_stat)
    print '    corrected ong --> ss P = %s' % str(
        3 * ong_ss_p_val), '(stat = %s)' % str(ong_ss_stat)

    y_offset = 0.004
    y1 = 0.42
    y2 = 0.48
    y_min = -0.101
    y_max = 0.501
    loc = plticker.MultipleLocator(base=0.1)

    ### draw one line connecting each pair of epochs for the population ###
    ### line alpha and asterisks indicate Wilcoxon signed-rank p-val ###

    if ong_trans_p_val < 0.05 / 3.:
        if 0.001 / 3. <= ong_trans_p_val < 0.01 / 3.:  #corrected for multiple comparisons
            ong_trans_p_for_plot = '**'
        elif ong_trans_p_val < 0.001 / 3.:
            ong_trans_p_for_plot = '***'
        else:
            ong_trans_p_for_plot = '*'
        ax.plot([-0.1, 0.75], [y1, y1], color='k', lw=2.0)

    else:
        x_pos = 0.325
        ong_trans_p_for_plot = ''
        ax.plot([-0.1, 0.75], [y1, y1], color='k', alpha=0.35, lw=2.0)

    x_pos = 0.325
    ax.text(x_pos,
            y1 + y_offset,
            ong_trans_p_for_plot,
            fontsize=8,
            horizontalalignment='center')

    if trans_ss_p_val < 0.05 / 3.:
        if 0.001 / 3. <= trans_ss_p_val < 0.01 / 3.:
            trans_ss_p_for_plot = '**'
        elif trans_ss_p_val < 0.001 / 3.:
            trans_ss_p_for_plot = '***'
        else:
            trans_ss_p_for_plot = '*'

        ax.plot([1.3, 2.1], [y1, y1], color='k', lw=2.0)

    else:
        trans_ss_p_for_plot = ''
        ax.plot([1.3, 2.1], [y1, y1], color='k', alpha=0.35, lw=2.0)

    x_pos = 1.7
    ax.text(x_pos,
            y1 + y_offset,
            trans_ss_p_for_plot,
            fontsize=8,
            horizontalalignment='center')

    if ong_ss_p_val < 0.05 / 3.:
        if 0.001 / 3. <= ong_ss_p_val < 0.01 / 3.:
            ong_ss_p_for_plot = '**'
        elif ong_ss_p_val < 0.001 / 3.:
            ong_ss_p_for_plot = '***'
        else:
            ong_ss_p_for_plot = '*'
        ax.plot([-0.1, 2.1], [y2, y2], color='k', lw=2.0)

    else:
        x_pos = 1.0
        ong_ss_p_for_plot = ''
        ax.plot([-0.1, 2.1], [y2, y2], color='k', alpha=0.35, lw=2.0)

    x_pos = 1.0
    ax.text(x_pos,
            y2 + y_offset,
            ong_ss_p_for_plot,
            fontsize=8,
            horizontalalignment='center')

    ### draw lines connecting markers across epochs for each pair,
    ### with line style determined by significance of change across
    ### epoch transition

    for i, ent in enumerate(trans_boots):
        trans_ong_overlap = False
        for entry in trans_boots[i]:
            if ong_boots[i][0] <= entry <= ong_boots[i][1]:
                trans_ong_overlap = True
                break
        for entry in ong_boots[i]:
            if trans_boots[i][0] <= entry <= trans_boots[i][1]:
                trans_ong_overlap = True
                break
        if trans_ong_overlap:
            ls = '-'
            alpha = 0.4
            lw = 0.7
        else:
            ls = '-'
            alpha = 1.0
            lw = 1.0
        ax.plot([0 + jitters[i], 1 + jitters[i]], [ong_CCs[i], trans_CCs[i]],
                color='k',
                ls=ls,
                lw=lw,
                alpha=alpha)

        ss_trans_overlap = False
        for entry in ss_boots[i]:
            if trans_boots[i][0] <= entry <= trans_boots[i][1]:
                ss_trans_overlap = True
                break
        for entry in trans_boots[i]:
            if ss_boots[i][0] <= entry <= ss_boots[i][1]:
                ss_trans_overlap = True
                break
        if ss_trans_overlap:
            ls = '-'
            alpha = 0.4
            lw = 0.7
        else:
            ls = '-'
            alpha = 1.0
            lw = 1.0
        ax.plot([1 + jitters[i], 2 + jitters[i]], [trans_CCs[i], ss_CCs[i]],
                color='k',
                ls=ls,
                lw=lw,
                alpha=alpha)

    ### Plot points for each pair (with fill based on significance relative
    ### to zero).

    #ongoing
    for q, entry in enumerate(ong_CCs):
        ong_zero_overlap = False
        if ong_boots[q][0] <= 0 <= ong_boots[q][1]:
            ong_zero_overlap = True
        if ong_zero_overlap:
            ong_mfc = (1, 1, 1)
            ong_alpha = 1.0
        else:
            ong_mfc = y
            ong_alpha = 1.0
        ax.plot(0 + jitters[q],
                ong_CCs[q],
                'o',
                mec='k',
                mfc=ong_mfc,
                mew=0.5,
                alpha=ong_alpha,
                markersize=markersize,
                clip_on=False)

    #transient
    for c, val in enumerate(trans_CCs):
        trans_zero_overlap = False
        if trans_boots[c][0] <= 0 <= trans_boots[c][1]:
            trans_zero_overlap = True
        if trans_zero_overlap:
            trans_mfc = (1, 1, 1)
            trans_alpha = 1.0
        else:
            trans_mfc = b
            trans_alpha = 1.0
        ax.plot(1 + jitters[c],
                trans_CCs[c],
                'o',
                mec='k',
                mfc=trans_mfc,
                mew=0.5,
                alpha=trans_alpha,
                markersize=markersize,
                clip_on=False)

    #steady-state
    for d, val in enumerate(ss_CCs):
        ss_zero_overlap = False
        if ss_boots[d][0] <= 0 <= ss_boots[d][1]:
            ss_zero_overlap = True
        if ss_zero_overlap:
            ss_mfc = (1, 1, 1)
            ss_alpha = 1.0
        else:
            ss_mfc = g
            ss_alpha = 1.0
        ax.plot(2 + jitters[d],
                ss_CCs[d],
                'o',
                mec='k',
                mfc=ss_mfc,
                mew=0.5,
                alpha=ss_alpha,
                markersize=markersize,
                clip_on=False)

    ### configure the plot ###
    ax.set_xlim(-0.2, 2.2)
    ax.set_ylim(y_min, y_max)
    #    y_labels_ = ax.get_yticks()
    #    y_labels = []
    #
    #    for h, label in enumerate(y_labels_):
    #        if (h-1)%3 == 0:
    #            y_labels.append(label)
    #        else:
    #            y_labels.append('')
    #    ax.set_yticklabels(y_labels, fontsize = size)
    #labels = ['-0.1', '', '0.1', '', '0.3']
    ax.yaxis.set_major_locator(loc)
    #ax.set_yticklabels(labels, size = size)

    #ax.locator_params(axis = 'y', nbins = 4)

    #plt.ylabel('<CC(0)>', fontsize = 14)
    ax.set_xticks([0, 1, 2])
    labels = ['ongoing', 'transient', 'steady-\nstate']
    #labels = ['', '', '']
    ax.set_xticklabels(labels, size=size)
    ax.set_yticklabels(ax.get_yticks(), size=size)

    ax.tick_params('both', length=5.5, width=1.3, which='major')
    ax.spines['top'].set_visible(False)
    ax.spines['right'].set_visible(False)
    ax.spines['bottom'].set_visible(False)
    ax.xaxis.set_ticks_position('none')
    ax.yaxis.set_ticks_position('left')
    y_label = 'CC ('
    if type(crit_freq) == float:
        y_label += str(int(crit_freq)) + ' Hz)'
    else:
        y_label += str(int(crit_freq[0])) + '-' + str(int(
            crit_freq[1])) + ' Hz)'
    ax.set_ylabel(y_label, fontsize=size)
Exemplo n.º 2
0
def plot_trajectory(ax, fontproperties, V_dict, boots_dict,
                    freq_band_for_plot):
    '''
    Plot the trial-averaged membrane potentials in 'trajectory'
    form.  Each dot indicates the across-trial average membrane potential for
    one cell, for the indicated epoch. Lines connect dots for individual cells 
    across epochs.  Lines have higher opacity and lineweight if change is 
    significant (i.e., bootstrap bands from the two epochs do not overlap).  
    Line and label across top indicate results of test for significant change 
    in population-average value across epochs (Wilcoxon signed-rank test, with 
    higher-opacity line indicating p < 0.05, and asterisks used to indicate 
    value of p if significant).
    
    Parameters
    ----------
    ax : matplotlib axis
    fontproperties: FontProperties object
        dictates size and font of text
    V_dict: python dictionary
        dictionary containing across-trial average V values for each cell,
        and bootstrap ranges.
    freq_band_for_plot: tuple of floats
        range of frequencies considered in Hz (e.g., (20.0, 100.0))
        
    Returns
    -------
    none    
    '''

    fonts = fontproperties
    size = fonts.get_size()
    markersize = 6

    y = (1, 1, 0)  #colors for dots
    g = (0, 1, 0)
    b = (0.2, 0.6, 1)

    ong_params = V_dict['ongoing']
    ong_boots = boots_dict['ongoing']
    trans_params = V_dict['transient']
    trans_boots = boots_dict['transient']
    ss_params = V_dict['steady-state']
    ss_boots = boots_dict['steady-state']

    #need left-right "jitter" to make overlapping dots distinguishable
    jitters = numpy.zeros(len(trans_params))
    for n, val in enumerate(jitters):
        jitters[n] += numpy.random.uniform(-0.05, 0.05)

    ### Get p-vals for stim modulation of population, draw on plot.
    ong_trans_stat, ong_trans_p_val = wlcx(ong_params, trans_params)
    trans_ss_stat, trans_ss_p_val = wlcx(trans_params, ss_params)
    ong_ss_stat, ong_ss_p_val = wlcx(ong_params, ss_params)

    #correct p-vals for multiple comparisons
    ong_trans_p_val *= 1. / 3.
    trans_ss_p_val *= 1. / 3.
    ong_ss_p_val *= 1. / 3.

    print '  <V> ong = %s' % str(
        numpy.mean(ong_params)), '+/- %s (mV; mean +/- s.e.m.)' % str(
            numpy.std(ong_params))
    print '  <V> trans = %s' % str(
        numpy.mean(trans_params)), '+/- %s (mV; mean +/- s.e.m.)' % str(
            numpy.std(trans_params))
    print '  <V> ss = %s' % str(
        numpy.mean(ss_params)), '+/- %s (mV; mean +/- s.e.m.)' % str(
            numpy.std(ss_params))
    print '  ong --> trans P_corrected = %s' % str(
        ong_trans_p_val), '(stat = %s)' % str(ong_trans_stat)
    print '  trans --> ss P_corrected = %s' % str(
        trans_ss_p_val), '(stat = %s)' % str(trans_ss_stat)
    print '  ong --> ss P_corrected = %s' % str(
        ong_ss_p_val), '(stat = %s)' % str(ong_ss_stat)

    y_offset = 0.5
    y1 = max(max(ong_params), max(trans_params), max(ss_params)) + 0.02
    y2 = y1 + 3.5

    if ong_trans_p_val < 0.05:
        if 0.001 <= ong_trans_p_val < 0.01:
            ong_trans_p_for_plot = '**'
        elif ong_trans_p_val < 0.001:
            ong_trans_p_for_plot = '***'
        else:
            ong_trans_p_for_plot = '*'
        ax.plot([-0.1, 0.75], [y1, y1], color='k', lw=2.0)

    else:
        x_pos = 0.325
        ong_trans_p_for_plot = ''
        ax.plot([-0.1, 0.75], [y1, y1], color='k', alpha=0.35, lw=2.0)

    x_pos = 0.325
    ax.text(x_pos,
            y1 + y_offset,
            ong_trans_p_for_plot,
            fontsize=size,
            horizontalalignment='center')

    if trans_ss_p_val < 0.05:
        if 0.001 <= trans_ss_p_val < 0.01:
            trans_ss_p_for_plot = '**'
        elif trans_ss_p_val < 0.001:
            trans_ss_p_for_plot = '***'
        else:
            trans_ss_p_for_plot = '*'

        ax.plot([1.3, 2.1], [y1, y1], color='k', lw=2.0)

    else:
        trans_ss_p_for_plot = ''
        ax.plot([1.3, 2.1], [y1, y1], color='k', alpha=0.35, lw=2.0)

    x_pos = 1.7
    ax.text(x_pos,
            y1 + y_offset,
            trans_ss_p_for_plot,
            fontsize=size,
            horizontalalignment='center')

    if ong_ss_p_val < 0.05:
        if 0.001 <= ong_ss_p_val < 0.01:
            ong_ss_p_for_plot = '**'
        elif ong_ss_p_val < 0.001:
            ong_ss_p_for_plot = '***'
        else:
            ong_ss_p_for_plot = '*'
        ax.plot([-0.1, 2.1], [y2, y2], color='k', lw=2.0)

    else:
        x_pos = 1.0
        ong_ss_p_for_plot = ''
        ax.plot([-0.1, 2.1], [y2, y2], color='k', alpha=0.35, lw=2.0)

    x_pos = 1.0
    ax.text(x_pos,
            y2 + y_offset,
            ong_ss_p_for_plot,
            fontsize=size,
            horizontalalignment='center')

    ### draw lines connecting markers across epochs for each pair,
    ### with line style determined by significance of change across
    ### epoch transition

    for i, ent in enumerate(trans_boots):
        trans_ong_overlap = False
        for entry in trans_boots[i]:
            if ong_boots[i][0] <= entry <= ong_boots[i][1]:
                trans_ong_overlap = True
                break
        for entry in ong_boots[i]:
            if trans_boots[i][0] <= entry <= trans_boots[i][1]:
                trans_ong_overlap = True
                break
        if trans_ong_overlap:
            ls = '-'
            alpha = 0.4
            lw = 0.7
        else:
            ls = '-'
            alpha = 1.0
            lw = 1.0
        ax.plot([0 + jitters[i], 1 + jitters[i]],
                [ong_params[i], trans_params[i]],
                color='k',
                ls=ls,
                lw=lw,
                alpha=alpha)

        ss_trans_overlap = False
        for entry in ss_boots[i]:
            if trans_boots[i][0] <= entry <= trans_boots[i][1]:
                ss_trans_overlap = True
                break
        for entry in trans_boots[i]:
            if ss_boots[i][0] <= entry <= ss_boots[i][1]:
                ss_trans_overlap = True
                break
        if ss_trans_overlap:
            ls = '-'
            alpha = 0.4
            lw = 0.7
        else:
            ls = '-'
            alpha = 1.0
            lw = 1.0
        ax.plot([1 + jitters[i], 2 + jitters[i]],
                [trans_params[i], ss_params[i]],
                color='k',
                ls=ls,
                lw=lw,
                alpha=alpha)

    ### Plot points for each pair ### .

    #ongoing
    for q, entry in enumerate(ong_params):
        ong_mfc = y
        ong_alpha = 1.0
        ax.plot(0 + jitters[q],
                ong_params[q],
                'o',
                mec='k',
                mfc=ong_mfc,
                mew=0.5,
                alpha=ong_alpha,
                markersize=markersize,
                clip_on=False)

    #transient
    for c, val in enumerate(trans_params):
        trans_mfc = b
        trans_alpha = 1.0
        ax.plot(1 + jitters[c],
                trans_params[c],
                'o',
                mec='k',
                mfc=trans_mfc,
                mew=0.5,
                alpha=trans_alpha,
                markersize=markersize,
                clip_on=False)

    #steady-state
    for d, val in enumerate(ss_params):
        ss_mfc = g
        ss_alpha = 1.0
        ax.plot(2 + jitters[d],
                ss_params[d],
                'o',
                mec='k',
                mfc=ss_mfc,
                mew=0.5,
                alpha=ss_alpha,
                markersize=markersize,
                clip_on=False)

    ### configure the plot ###
    ax.set_xlim(-0.2, 2.2)
    loc = plticker.MultipleLocator(base=20)
    ax.yaxis.set_major_locator(loc)
    #ax.locator_params(axis = 'y', nbins = 4)

    ax.set_xticks([0, 1, 2])
    labels = ['ongoing', 'transient', 'steady-\nstate']
    ax.set_xticklabels(labels, size=size)
    for tick in ax.yaxis.get_major_ticks():
        tick.label.set_fontsize(size)
    #ax.tick_params('both', length = 5.5, width = 1.3, which = 'major')
    ax.spines['top'].set_visible(False)
    ax.spines['right'].set_visible(False)
    ax.spines['bottom'].set_visible(False)
    ax.xaxis.set_ticks_position('none')
    ax.yaxis.set_ticks_position('left')
    yaxis_label = r'$\overline{V}$ (mV)'
    ax.set_ylabel(yaxis_label, fontsize=size)
Exemplo n.º 3
0
def plot_trajectory(ax, fontproperties, rP_dict, boots_dict,
                    freq_band_for_plot):
    '''
    Plot the trial-averaged relative power (evoked/ongoing) in 'trajectory'
    form.  Each dot indicates the across-trial average relative power for
    one cell, for the indicated evoked epoch.  Dot is filled if value is
    significant (i.e., bootstrap bands do not overlap rP = 1), empty otherwise.
    Lines connect dots for individual cells across epochs.  Lines have higher
    opacity and lineweight if change is significant (i.e., bootstrap bands from
    the two epochs do not overlap).  Line and label across top indicate results
    of test for significant change in population-average value across epochs
    (Wilcoxon signed-rank test, with higher-opacity line indicating p < 0.05,
    and asterisks used to indicate value of p if significant).
    
    Parameters
    ----------
    ax : matplotlib axis
    fontproperties: FontProperties object
        dictates size and font of text
    rP_dict: python dictionary
        dictionary containing across-trial average rP values for each cell,
        and bootstrap ranges.
    freq_band_for_plot: tuple of floats
        range of frequencies considered in Hz (e.g., (20.0, 100.0))
        
    Returns
    -------
    none    
    '''

    fonts = fontproperties
    size = fonts.get_size()
    mpl.rcParams['mathtext.default'] = 'regular'

    if type(freq_band_for_plot) == float:
        freq_band_label = 'rP (0 - %s Hz)' % str(freq_band_for_plot)
    else:
        freq_band_label = r'$\ rP_{hf}$ (%s - ' % str(
            int(freq_band_for_plot[0])) + '%s Hz)' % str(
                int(freq_band_for_plot[1]))
    y_offset = 2
    y1 = 190
    ax.set_ylabel(freq_band_label, fontsize=size)

    trans_params = rP_dict['transient']
    trans_boots = boots_dict['transient']
    ss_params = rP_dict['steady-state']
    ss_boots = boots_dict['steady-state']

    #need left-right "jitter" to make overlapping dots distinguishable
    jitters = numpy.zeros(len(trans_params))
    for n, val in enumerate(jitters):
        jitters[n] += numpy.random.uniform(-0.05, 0.05)

    ### Get p-vals for stim modulation of population, draw on plot.

    trans_ss_stat, trans_ss_p_val = wlcx(trans_params, ss_params)

    print '  <rP> trans = %s' % str(
        numpy.mean(trans_params)), '+/- %s (mV; mean +/- s.e.m.)' % str(
            numpy.std(trans_params))
    print '    (P = %s; one-sided t-test)' % str(
        0.5 * ttest_1samp(trans_params, 0)[1])
    print '  <rP> ss = %s' % str(
        numpy.mean(ss_params)), '+/- %s (mV; mean +/- s.e.m.)' % str(
            numpy.std(ss_params))
    print '    (P = %s; one-sided t-test)' % str(
        0.5 * ttest_1samp(ss_params, 0)[1])
    print '  trans --> ss P = %s' % str(trans_ss_p_val), '(stat = %s)' % str(
        trans_ss_stat)

    if trans_ss_p_val < 0.05:
        if 0.001 <= trans_ss_p_val < 0.01:
            trans_ss_p_for_plot = '**'
        elif trans_ss_p_val < 0.001:
            trans_ss_p_for_plot = '***'
        else:
            trans_ss_p_for_plot = '*'

        ax.plot([0.1, 0.9], [y1, y1], color='k', lw=1.5)

    else:
        trans_ss_p_for_plot = ''
        ax.plot([0.1, 0.9], [y1, y1], color='k', alpha=0.35, lw=1.5)

    x_pos = 0.5
    ax.text(x_pos,
            y1 + y_offset,
            trans_ss_p_for_plot,
            fontsize=8,
            horizontalalignment='center')

    for n, val in enumerate(trans_params):
        ### truncate one outlier value for the plot ###
        if trans_params[n] > 200:
            print '    truncated the value rP = %s (transient epoch) for plot' % str(
                trans_params[n])
            trans_params[n] += 200 - val

    ### draw lines connecting markers across epochs for each trode,
    ### with line style determined by significance of change across
    ### epoch transition

    for i, ent in enumerate(trans_boots):
        ss_trans_overlap = False
        for entry in ss_boots[i]:
            if trans_boots[i][0] <= entry <= trans_boots[i][1]:
                ss_trans_overlap = True
                break
        for entry in trans_boots[i]:
            if ss_boots[i][0] <= entry <= ss_boots[i][1]:
                ss_trans_overlap = True
                break
        if ss_trans_overlap:
            ls = '-'
            alpha = 0.4
            lw = 0.4
        else:
            ls = '-'
            alpha = 0.7
            lw = 0.7
        ax.plot([0 + jitters[i], 1 + jitters[i]],
                [trans_params[i], ss_params[i]],
                color='k',
                ls=ls,
                lw=lw,
                alpha=alpha)

    ### Plot points for each trode (with fill based on significance relative
    ### to rP = 1.0).

    #transient
    for c, val in enumerate(trans_params):
        trans_one_overlap = False
        if trans_boots[c][0] <= 1.0 <= trans_boots[c][1]:
            trans_one_overlap = True
        if trans_one_overlap:
            trans_mfc = (1, 1, 1)
            trans_alpha = 1.0
        else:
            trans_mfc = b
            trans_alpha = 1.0
        ax.plot(0 + jitters[c],
                trans_params[c],
                'o',
                mec='k',
                mfc=trans_mfc,
                mew=0.75,
                alpha=trans_alpha,
                markersize=6,
                clip_on=False)

    #steady-state
    for d, val in enumerate(ss_params):
        ss_one_overlap = False
        if ss_boots[d][0] <= 1.0 <= ss_boots[d][1]:
            ss_one_overlap = True
        if ss_one_overlap:
            ss_mfc = (1, 1, 1)
            ss_alpha = 1.0
        else:
            ss_mfc = g
            ss_alpha = 1.0
        ax.plot(1 + jitters[d],
                ss_params[d],
                'o',
                mec='k',
                mfc=ss_mfc,
                mew=0.75,
                alpha=ss_alpha,
                markersize=6,
                clip_on=False)

    ### configure the plot ###
    ax.set_xlim(-0.2, 1.2)
    ax.set_xticks([0, 1])
    x_labels = ['transient', 'steady-\nstate']
    ax.set_xticklabels(x_labels, size=size)

    ### Warning!!! These axis labels are set manually for the final publication
    ### figure.  If parameters are changed, this section should be
    ### commented out.
    y_labels = ['0', '50', '100', '150', '>200']
    ax.set_yticklabels(y_labels, rotation='vertical', size=size)

    ax.tick_params('both', length=4, width=1.0, which='major')
    ax.spines['top'].set_visible(False)
    ax.spines['right'].set_visible(False)
    ax.spines['bottom'].set_visible(False)
    ax.xaxis.set_ticks_position('none')
    ax.yaxis.set_ticks_position('left')
Exemplo n.º 4
0
def report_CC_results_model_network(
        node_group='1_center',
        N_E=800,
        N_I=200,
        tau_Em=50,
        tau_Im=25,
        p_exc=0.03,
        p_inh=0.20,
        tau_ref_E=10,
        tau_ref_I=5,
        V_leak_E_min=-70,
        V_leak_I_min=-70,
        V_th=-40,
        V_reset=-59,
        V_peak=0,
        V_syn_E=50,
        V_syn_I=-68,
        C_E=0.4,
        C_I=0.2,
        g_AE=3.0e-3,
        g_AI=6.0e-3,
        g_GE=30.0e-3,
        g_GI=30.0e-3,
        g_extE=6.0e-3,
        g_extI=6.0e-3,
        g_leak_E=10.0e-3,
        g_leak_I=5.0e-3,
        tau_Fl=1.5,
        tau_AIl=1.5,
        tau_AIr=0.2,
        tau_AId=1.0,
        tau_AEl=1.5,
        tau_AEr=0.2,
        tau_AEd=1.0,
        tauR_intra=50000.0,
        tauD_intra=300.0,
        tauR_extra=50000.0,
        tauD_extra=300.0,
        rateI=65,
        rateE=65,
        stim_factor=4.5,
        num_nodes_to_save=20,
        num_pairs=40,
        num_LFPs=8,
        windows=[1000.0, 1000.0, 1000.0],
        gaps=[200.0, 0.0, 200.0],
        padding=500.0,
        stim_type='linear_increase',
        time_to_max=200,
        exc_connectivity='clustered',
        inh_connectivity='random',
        exc_syn_weight_dist='beta',
        net_state='synchronous',
        ext_SD=True,
        int_SD=True,
        generate_LFPs=True,
        save_results=True,
        num_trials=15,
        crit_freq=(20.0, 100.0),
        filt_kind='band',
        run_new_net_sim=False,
        get_new_CC_results_all_pairs=False,
        master_folder_path='F:\\correlated_variability'):
    '''Calculate CC values for each epoch for pairs of 'test' neurons.  'Test' 
    neurons receive synaptic inputs generated by the model network.  

    Calculate the trial-averaged Pearson correlation coefficient (CC) for each
    test pair, for the ongoing, transient, and steady-state epochs, 
    for the crit_freq freq band.  For each epoch, report mean +/- s.e.m.  
    For the population, the significance of the across-epoch change is 
    assessed via the Wilcoxon signed-rank test.  Results are printed in the
    command line.  This function is similar to 
    sub_fig_plotting.plot_CC_trajectories_model_neurons, but generates no
    figures.  It can be useful when exploring model parameters.
    
    
    Parameters
    ----------
    ax: matplotlib axis
    fontproperties: FontProperties object
        dictates size and font of text
    node_group: float
        group of network nodes from which to select example Vs
    tau_Em: float
        membrane time constant in ms for exc nodes
    tau_Im: float
        membrane time constant in ms for inh nodes
    N_E: float or int
        number of exc nodes
    N_I: float or int
        number of inh nodes
    p_exc: float
        connection probability for E --> E and E --> I
    p_inh: float
        connection probability for I --> E and I --> I
    tau_ref_E: float
        absolute refractory period for exc nodes in ms
    tau_ref_I: float
        absolute refractory period for inh nodes in ms
    V_leak_I: float or int
        leak reversal potential for inh nodes in mV
    V_leak_E: float or int
        leak reversal potential for exc nodes in mV
    V_th: float or int
        spike threshold for all nodes in mV
    V_reset: float or int
        post-spike reset membrane potential for all nodes in mV
    V_peak: float or int
        peak spike membrane potential for all nodes in mV
    dt: float
        time step in ms
    V_syn_E: float or int
        synaptic reversal potential for exc nodes in mV
    V_syn_I: float or int
        synaptic reversal potential for inh nodes in mV
    g_AE: float
        synaptic conductance for AMPA channels on exc nodes in microS
    g_AI: float 
        synaptic conductance for AMPA channels on inh nodes in microS
    g_GE: float 
        synaptic conductance for GABA channels on exc nodes in microS
    g_GI: float 
        synaptic conductance for GABA channels on inh nodes in microS
    tau_Fl: float or in f
        inh to inh delay time constant in ms
    tau_Fr: float
        inh to inh rise time constant in ms
    tau_Fd: float
        inh to inh decay time constant in ms
    tau_AIl: float
        exc to inh AMPA delay time constant in ms
    tau_AIr: float
        exc to inh AMPA rise time constant in ms
    tau_AId: float 
        exc to inh AMPA decay time constant in ms
    tau_AEl: float
        exc to exc AMPA delay time constant in ms
    tau_AEr: float
        exc to exc AMPA rise time constant in ms
    tau_AEd: float 
        exc to exc AMPA decay time constant in ms
    g_extE: float
        synaptic conductance for ext input on exc in microS
    g_extI: float
        synaptic conductance for ext input on inh in microS
    C_E: float 
        capacitance for exc nodes in nF
    C_I: float 
        capacitance for inh nodes in nF
    g_leak_E: float
        leak conductance for exc nodes in microS
    g_leak_I: float l
        leak conductance for inh nodes in microS
    rateI: float or int 
        ongoing external input rate for exc nodes in Hz
    rateE: float or int 
        ongoing external input rate for inh nodes in Hz
    stim_factor: float or int
        factor by which external input increases after stim onset 
    num_nodes_to_save: int
        number of nodes for which to save synaptic inputs (for injecting
        into 'test' neurons in a separate function) 
    num_pairs: int
        number of test neuron pairs for which to calculate CCs
    num_LFPs: float
        number of LFPs to simulate (determines # of nodes in each 'electrode')
    windows: list of floats
        widths of ongoing, transient, and steady-state windows (ms)
    gaps: list of floats
        sizes of gaps between stim onset and end of ongoing, stim onset and
        beginning of transient, end of transient and beginning of 
        steady-state (ms)
    padding: float
        size of window (ms) to be added to the beginning and end of each 
        simulation
    stim_type: string
        if 'linear_increase', the stimulus is mimicked as a gradual (step-wise)
        increase in external input rate.  Otherwise, single step function.
    time_to_max: float or int
        time to reach max input rate for 'linear_increase' stimulus (in ms)
    exc_connectivity: string
        type of E --> E and E --> I connectivity ('clustered' or 'random')
    inh_connectivity: string
        type of I --> E and I --> I connectivity ('clustered' or 'random')
    exc_syn_weight_dist: string
        type of distribution from which to draw E --> E and E --> I nonzero
        weights.  If 'beta', draw from beta distribution.  Otherwise, draw
        from continuous uniform distribution.
    net_state: string
        If 'synchronous', choose inh synaptic time constants that support
        network spike-rate oscillations in response to the stimulus.
        Otherwise, use same time constants for inh and exc synapses.
    ext_SD: bool
        if True, apply synaptic adaptation to external input synapses after
        stimulus onset
    int_SD: bool
        if True, apply synaptic adaptation to all 'intracortical' synapses 
        after stimulus onset        
    generate_LFPs: bool
        if True, simulate LFPs as sums of synaptic currents to groups of
        exc nodes
    save_results: bool
        if True, save results for each trial
    num_trials: float or int
        number of trials to simulation
    crit_freq: float or tuple of floats 
        critical frequency for filtering membrane potentials 
        (e.g., (20.0, 100.0))
    filt_kind: string
        kind of filter to use for membrane potentials (e.g., 'band' for 
        bandpass) 
    run_new_net_sim: Bool
        if True, run new network simulations (for num_trials trials), even if
        results for these settings already exist
    get_new_CC_results_all_pairs: Bool
        if True, calculate CCs for all pairs starting with traces.
    master_folder_path: string
        full path of directory containing data, code, figures, etc.

    Returns
    -------
    None
        
    '''

    model_version = exc_connectivity + '_' + net_state
    if int_SD == False or ext_SD == False:
        model_version += '_no_SD'
    print ''
    print '  model version:', model_version
    print '  node group:', node_group

    CCs_all_pairs_and_trials = get_CC(
        node_group, N_E, N_I, tau_Em, tau_Im, p_exc, p_inh, tau_ref_E,
        tau_ref_I, V_leak_E_min, V_leak_I_min, V_th, V_reset, V_peak, V_syn_E,
        V_syn_I, C_E, C_I, g_AE, g_AI, g_GE, g_GI, g_extE, g_extI, g_leak_E,
        g_leak_I, tau_Fl, tau_AIl, tau_AIr, tau_AId, tau_AEl, tau_AEr, tau_AEd,
        tauR_intra, tauD_intra, tauR_extra, tauD_extra, rateI, rateE,
        stim_factor, num_nodes_to_save, num_pairs, num_LFPs, windows, gaps,
        padding, stim_type, time_to_max, exc_connectivity, inh_connectivity,
        exc_syn_weight_dist, net_state, ext_SD, int_SD, generate_LFPs,
        save_results, num_trials, crit_freq, filt_kind, run_new_net_sim,
        get_new_CC_results_all_pairs, master_folder_path)

    epochs = ['ongoing', 'transient', 'steady-state']
    CC_dict = {}

    for epoch in epochs:
        CC_dict[epoch] = []
        for pair in CCs_all_pairs_and_trials:
            CCs_all_trials = CCs_all_pairs_and_trials[pair]
            CC_dict[epoch].append(numpy.mean(CCs_all_trials[epoch]))

    ong_CCs = CC_dict['ongoing']
    trans_CCs = CC_dict['transient']
    ss_CCs = CC_dict['steady-state']

    ong_trans_p_val = wlcx(ong_CCs, trans_CCs)[1]
    trans_ss_p_val = wlcx(trans_CCs, ss_CCs)[1]
    ong_ss_p_val = wlcx(ong_CCs, ss_CCs)[1]
    ong_trans_stat = wlcx(ong_CCs, trans_CCs)[0]
    trans_ss_stat = wlcx(trans_CCs, ss_CCs)[0]
    ong_ss_stat = wlcx(ong_CCs, ss_CCs)[0]

    print '    <CC> ong = %s' % str(
        numpy.mean(ong_CCs)), '+/-', numpy.std(ong_CCs)
    print '      (P = %s; one-sided t-test)' % str(
        0.5 * ttest_1samp(ong_CCs, 0)[1])
    print '    <CC> trans = %s' % str(
        numpy.mean(trans_CCs)), '+/-', numpy.std(trans_CCs)
    print '      (P = %s; one-sided t-test)' % str(
        0.5 * ttest_1samp(trans_CCs, 0)[1])
    print '    <CC> ss = %s' % str(
        numpy.mean(ss_CCs)), '+/-', numpy.std(ss_CCs)
    print '      (P = %s; one-sided t-test)' % str(
        0.5 * ttest_1samp(ss_CCs, 0)[1])

    print '    ong --> trans P = %s' % str(
        ong_trans_p_val), '(stat = %s)' % str(ong_trans_stat)
    print '    trans --> ss P = %s' % str(trans_ss_p_val), '(stat = %s)' % str(
        trans_ss_stat)
    print '    ong --> ss P = %s' % str(ong_ss_p_val), '(stat = %s)' % str(
        ong_ss_stat)

    print '    corrected ong --> trans P = %s' % str(
        3 * ong_trans_p_val), '(stat = %s)' % str(ong_trans_stat)
    print '    corrected trans --> ss P = %s' % str(
        3 * trans_ss_p_val), '(stat = %s)' % str(trans_ss_stat)
    print '    corrected ong --> ss P = %s' % str(
        3 * ong_ss_p_val), '(stat = %s)' % str(ong_ss_stat)
Exemplo n.º 5
0
Description: Wilcoxon signed ranked test.
"""
from scipy.stats import wilcoxon as wlcx
import itertools as it

alpha = 0.05
#n = len(X1)
crit_val = 25  #use table

##### Sample Data #####
# X1 = [70.14,72.12,71.28,72.35,66.67,64.81,69.17,68.18,72.27,71.29,75.63,72.85,72.88,69.76,65]
# X2 = [66.77,71.29,65.59,69.24,66.34,62.22,49.45,65.91,64.77,70.61,63.51,70.52,77.58,63.7,63.33]
# X3 = [68.86,69.02,69.76,71.59,69.26,62.13,69.09,66.67,65.3,68.56,73.36,70.64,70.61,68.43,67.5]
# X4 = [69.37,79.17,72.73,71.52,70.74,63.89,69.92,68.94,72.35,72.12,72.66,71.91,79.02,70.45,67.5]

results_dict = dict(X1=X1, X2=X2, X3=X3, X4=X4)

result_list = list(map(dict, it.combinations(
    results_dict.items(), 2)))  #map results list into combinatorial pairs.

#####Iterate through all combinations of results#####
for sub_list in result_list:
    X1 = sub[list(sub_list.keys())[0]]
    X2 = sub[list(sub_list.keys())[1]]

    w_stat, p_val = wlcx(X1, X2)
    if w_stat < crit_val:
        print(f"Differences are significant with p<{p_val}")
    else:
        print(f"Differences are not significant! p value = {p_val}")