def plot_probheatmaps(probabilityArray,
fsds,
skip,
threshold=0.5,
smoothing=20,
channel1=0,
channel2=1):
fig, ax = plt.subplots(3, 2, figsize=(10, 10), dpi=300)
windows, nchan = probabilityArray.shape
thresholded = copy.deepcopy(probabilityArray)
thresholded[thresholded > threshold] = 1
thresholded[thresholded <= threshold] = 0
w = int(smoothing * fsds / skip)
mvgAvg = np.zeros(shape=(windows - w + 1, nchan))
for c in range(nchan):
mvgAvg[:, c] = echobase.movingaverage(probabilityArray[:, c], w)
mvgAvgThreshold = copy.deepcopy(mvgAvg)
mvgAvgThreshold[mvgAvgThreshold > threshold] = 1
mvgAvgThreshold[mvgAvgThreshold <= threshold] = 0
sns.lineplot(x=range(windows),
y=probabilityArray[:, channel1],
ci=None,
ax=ax[0, 0])
sns.lineplot(x=range(windows),
y=probabilityArray[:, channel2],
ci=None,
ax=ax[0, 1])
sns.heatmap(probabilityArray.T, ax=ax[1, 0])
sns.heatmap(thresholded.T, ax=ax[1, 1])
sns.heatmap(mvgAvg.T, ax=ax[2, 0])
sns.heatmap(mvgAvgThreshold.T, ax=ax[2, 1])
def plot_heatmapSingleThreshold(probabilityArray,
fsds,
skip,
figsize=(5, 2.5),
threshold=0.5,
smoothing=20,
vmax=1.1,
cmapName="Blues"):
fig, ax = plt.subplots(1, 1, figsize=figsize, dpi=300)
windows, nchan = probabilityArray.shape
thresholded = copy.deepcopy(probabilityArray)
thresholded[thresholded > threshold] = 1
thresholded[thresholded <= threshold] = 0
w = int(smoothing * fsds / skip)
mvgAvg = np.zeros(shape=(windows - w + 1, nchan))
for c in range(nchan):
mvgAvg[:, c] = echobase.movingaverage(probabilityArray[:, c], w)
mvgAvgThreshold = copy.deepcopy(mvgAvg)
mvgAvgThreshold[mvgAvgThreshold > threshold] = 1
mvgAvgThreshold[mvgAvgThreshold <= threshold] = 0
cmap = plt.cm.get_cmap(cmapName)
sns.heatmap(mvgAvgThreshold.T,
cmap=cmap,
ax=ax,
cbar=False,
xticklabels=False,
yticklabels=False,
vmax=vmax)
def plot_singleProbability(probabilityArray,
fsds,
skip,
channel=1,
startInd=0,
stopInd=None,
smoothing=20,
vmin=-0.2,
vmax=1.2):
windows, nchan = probabilityArray.shape
if stopInd == None:
stopInd = windows
smoothing = smoothing
w = int(smoothing * fsds / skip)
mvgAvg = np.zeros(shape=(windows - w + 1, nchan))
for c in range(nchan):
mvgAvg[:, c] = echobase.movingaverage(probabilityArray[:, c], w)
y = mvgAvg[startInd:stopInd, channel]
x = range(len(y))
y2 = -0.05
fig, ax = plt.subplots(1, 1, figsize=(10, 2), dpi=300)
sns.lineplot(x=x, y=y, ci=None, ax=ax, alpha=0)
interpolate = interp1d(np.array(x), np.array(x))
xnew = interpolate(np.arange(0, len(x) - 1, 1))
interpolate = interp1d(np.array(x), y)
ynew = interpolate(np.arange(0, len(x) - 1, 1))
Z = np.linspace(0, 1, len(xnew))
normalize = colors.Normalize(vmin=vmin, vmax=vmax)
cmap = plt.cm.get_cmap('Blues')
df = pd.DataFrame({'x': xnew, 'y1': ynew, 'z': Z})
xcolor = df['x'].values
y1color = df['y1'].values
zcolor = df['z'].values
for ii in range(len(df['x'].values) - 1):
plt.fill_between(x=[xcolor[ii], xcolor[(ii + 1)]],
y1=[y1color[ii], y1color[ii + 1]],
y2=y2,
color=cmap(normalize(zcolor[ii])))
ax.set(yticks=[])
ax.set(xticks=[])
ax.set_ylim([y2, 1])
sns.despine(bottom=True, left=True)
probWN = np.zeros(shape=(windows, len(c)))
probCNN = np.zeros(shape=(windows, len(c)))
probLSTM = np.zeros(shape=(windows, len(c)))
for ch in range(len(c)):
print(ch)
ch_pred = data_scalerDSPlot_X[:, :, ch].reshape(windows, time_step, 1)
probWN[:, ch] = modelWN.predict(ch_pred, verbose=1)[:, 1]
probCNN[:, ch] = modelCNN.predict(ch_pred, verbose=1)[:, 1]
probLSTM[:, ch] = modelLSTM.predict(ch_pred, verbose=1)[:, 1]
smoothing = 20
w = int(smoothing * fsds / skip)
mvgAvg = np.zeros(shape=(windows - w + 1, len(c)))
for ch in range(len(c)):
mvgAvg[:, ch] = echobase.movingaverage(probWN[:, ch], w)
#Calculate Raw Features
linelength, linelengthNorm = echobase.lineLengthOfArray(data_scalerDSPlot_X)
#%% generate example EEG
fnameFig = os.path.join(pathFigureOverview, "exampleEEGspread.png")
DataJson.plot_eeg(data_scaler[int(ictalStartIndex -
1 * fs_plot):int(ictalStopIndex +
2 * fs_plot), c],
fs_plot,
nchan=None,
dpi=300,
aspect=30,
height=0.5,
sns.heatmap(lineLength_arr.T, ax=ax[1, 0])
THRESHOLD_LL = 0.7
SMOOTHING = 20 #in seconds
lineLength_arr_threshold = copy.deepcopy(lineLength_arr)
lineLength_arr_threshold[lineLength_arr_threshold > THRESHOLD_LL] = 1
lineLength_arr_threshold[lineLength_arr_threshold <= THRESHOLD_LL] = 0
sns.heatmap(lineLength_arr_threshold.T, ax=ax[1, 1])
w = int(SMOOTHING * fsds / skip)
lineLength_arr_movingAvg = np.zeros(shape=(windows - w + 1, nchan))
for c in range(nchan):
lineLength_arr_movingAvg[:, c] = echobase.movingaverage(
lineLength_arr[:, c], w)
sns.heatmap(lineLength_arr_movingAvg.T, ax=ax[2, 0])
lineLength_arr_movingAvg_threshold = copy.deepcopy(lineLength_arr_movingAvg)
lineLength_arr_movingAvg_threshold[
lineLength_arr_movingAvg_threshold > THRESHOLD_LL] = 1
lineLength_arr_movingAvg_threshold[
lineLength_arr_movingAvg_threshold <= THRESHOLD_LL] = 0
sns.heatmap(lineLength_arr_movingAvg_threshold.T, ax=ax[2, 1])
#sns.heatmap(SC, square=True)
#sns.heatmap(SC_hat, square=True)