reportCallback

from keras.callbacks import Callback
import time
import matplotlib.pyplot as plt # conda install matplotlib
import matplotlib.lines as mlines

class ReportCallback(Callback):
    def __init__(self, filename, xTrain, yTrain):
        self.filename = filename
        self.report_file = open(filename+".csv", "w+")
        self.report_file.write("#epoch,time,loss,acc,val_loss,val_acc\n");
        self.allLoss = []
        self.allAcc = []
        self.allValLoss = []
        self.allValAcc = []
        self.xTrain = xTrain
        self.yTrain = yTrain

    def on_train_begin(self, logs={}):
        self.beginTime = time.time()

    def on_epoch_end(self, epoch, logs={}):
        self.allLoss.append(logs.get('loss'))
        self.allAcc.append(logs.get('acc'))
        self.allValLoss.append(logs.get('val_loss'))
        self.allValAcc.append(logs.get('val_acc'))

        self.report_file.write(
                        str(epoch+1)+","+
                        str(time.time()-self.beginTime)+","+
                        str(logs.get('loss'))+","+
                        str(logs.get('acc'))+","+
                        str(logs.get('val_loss'))+","+
                        str(logs.get('val_acc'))+"\n"
                        )

        if epoch+1 == self.params['nb_epoch']:
            self.report_file.close()

            x = range(1, epoch+2)

            fig, ax1 = plt.subplots()
            ax1.set_ylabel("Loss", color="#CC2529")
            ax1.plot(x, self.allLoss, "--", color="#CC2529", linewidth=2.0, label="Train Loss")
            ax1.plot(x, self.allValLoss, color="#CC2529", linewidth=2.0, label="Test Loss")
            ax1.axis('tight')

            ax2 = ax1.twinx()
            ax2.set_ylabel("Accuracy", color="#396BB1")
            ax2.plot(x, self.allAcc, "--", color="#396BB1", linewidth=2.0, label="Train Acc")
            ax2.plot(x, self.allValAcc, color="#396BB1", linewidth=2.0, label="Test Acc")
            ax2.axis('tight')

            ax1.set_xlabel("Epochs")
            training_line  = mlines.Line2D([], [], linestyle='--', linewidth=2.0, color="black", label="Training")
            test_line  = mlines.Line2D([], [], linewidth=2.0, color="black", label="Test")
            plt.legend(bbox_to_anchor=(0., 1.02, 1., .102), loc=3, ncol=2, mode="expand", borderaxespad=0., handles=[training_line, test_line])

            plt.savefig(self.filename+"_graph.png")