def evaluateCombinedTotalRocCurves(predictions, true, methods):
res = "{0: <20} {1: <19} {2: <19} {3: <19}\n".format(" ", "auc", "auc forward", "auc feedbacks")
plt.figure(figsize=(35, 12))
best_auc = 0
subplot_i, subplot_k = getSubplots(3)
for f in methods:
y_true, y_pred = true[f], predictions[f]
feedbacks_y_true = np.reshape([getFeedbackLinks(temp) for temp in y_true], (-1, 1))
feedbacks_y_pred = np.reshape([getFeedbackLinks(temp) for temp in y_pred], (-1, 1))
forward_y_true = np.reshape([getForwardLinks(temp) for temp in y_true], (-1, 1))
forward_y_pred = np.reshape([getForwardLinks(temp) for temp in y_pred], (-1, 1))
combined_y_pred = np.reshape(y_pred, (-1, 1))
combined_y_true = np.reshape(y_true, (-1, 1))
plt.subplot(subplot_i, subplot_k, 1)
roc_auc = plotROC(combined_y_true, combined_y_pred, f)
if roc_auc > best_auc and roc_auc < 0.99:
best_auc = roc_auc
plt.subplot(subplot_i, subplot_k, 2)
roc_auc_forward = plotROC(forward_y_true, forward_y_pred, f)
plt.title('ROC for forward only')
plt.subplot(subplot_i, subplot_k, 3)
roc_auc_feedbacks = plotROC(feedbacks_y_true, feedbacks_y_pred, f)
plt.title('ROC for feedbacks only')
res += "{0: <20} {1:16.3f} {2:16.3f} {3:16.3f}\n".format(f, roc_auc, roc_auc_forward, roc_auc_feedbacks)
plt.savefig(join(PROJECT_DIR, 'output', 'evaluation', s_timestamp() + '.pdf'))
with open(join(PROJECT_DIR, 'output', 'evaluation', s_timestamp() + '.txt'), 'w') as pout:
pout.write(res)
return best_auc
def plotPredictions(dataset, method, predict_n): cmap = plt.cm.Accent with PdfPages(join(PROJECT_DIR, 'output', 'visualization_graph_predictions_really', s_timestamp() + '.pdf')) as pdf: for idx_instance in range(dataset.n_instances): instance = dataset.get(idx_instance) plt.figure(figsize=(20, 14)) subplot_i, subplot_k = getSubplots(instance.n_time_series + 1) labels = ['Pulse', '1 Inhibition', '2 Inhibitions', 'Oscilatory', 'Oscilatory+1 Inhibition'] plt.subplot(subplot_i, subplot_k, 1) instance.plotDiGraphViaGraphViz_(cmap) for idx_time_series in range(instance.n_time_series): plt.subplot(subplot_i, subplot_k, idx_time_series + 2) instance.setx(idx_time_series) y_pred = method(instance) for idx_node in range(instance.n_nodes): y_pred[idx_node][idx_node] = 0 y_pred = y_pred.reshape(-1, ) plotPredicted(y_pred, labels[idx_time_series], predict_n, cmap, instance.n_nodes, instance.pos, instance.labels) pdf.savefig() # saves the current figure into a pdf page plt.close()
def evaluateIndividualRocCurvesAndPredictions(d, predictions, true, predict_n, methods):
with PdfPages(join(PROJECT_DIR, 'output', 'visualization_graph_predictions', s_timestamp() + '.pdf')) as pdf:
cmap = plt.cm.Accent
for idx_instance in range(d.n_instances):
instance = d.get(idx_instance)
n_nodes = instance.n_nodes
node_labels = instance.labels
plt.figure(figsize=(40, 20))
subplot_i, subplot_k = getSubplots(len(predictions) + 4)
for subplot_idx, f in enumerate(methods):
y_true = true[f][idx_instance][:]
y_pred = predictions[f][idx_instance][:]
# plot the roc curve for the instance
plt.subplot(subplot_i, subplot_k, len(predictions) + 1)
plotROC(y_true, y_pred, f)
# plot the roc curve for the feedbacks only
plt.subplot(subplot_i, subplot_k, len(predictions) + 2)
plotROC(getFeedbackLinks(y_true), getFeedbackLinks(y_pred), f)
plt.title('ROC for feedbacks only')
# plot the roc curve for the feedbacks only
plt.subplot(subplot_i, subplot_k, len(predictions) + 3)
plotROC(getForwardLinks(y_true), getForwardLinks(y_pred), f)
plt.title('ROC for forward only')
# plot the predicted networks
plt.subplot(subplot_i, subplot_k, subplot_idx + 1)
plotPredicted(y_pred, f, predict_n, cmap, n_nodes, instance.pos, node_labels)
plt.subplot(subplot_i, subplot_k, len(predictions) + 4)
instance.plotTimeSeries_(cmap)
pdf.savefig() # saves the current figure into a pdf page
plt.close()