#==============================================================================#
# Author: Dominik Müller #
# Copyright: 2020 IT-Infrastructure for Translational Medical Research, #
# University of Augsburg #
# #
# This program is free software: you can redistribute it and/or modify #
# it under the terms of the GNU General Public License as published by #
# the Free Software Foundation, either version 3 of the License, or #
# (at your option) any later version. #
# #
# This program is distributed in the hope that it will be useful, #
# but WITHOUT ANY WARRANTY; without even the implied warranty of #
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the #
# GNU General Public License for more details. #
# #
# You should have received a copy of the GNU General Public License #
# along with this program. If not, see <http://www.gnu.org/licenses/>. #
#==============================================================================#
# Import libraries
import pandas as pd
import os
from plotnine import *
# Hardcoded data pathes
path_all = "/home/mudomini/projects/covid19.MIScnn.RESULTS/evaluation/"
path_noDAnoPP = "/home/mudomini/projects/covid19.MIScnn.RESULTS/evaluation.stepwise.noDA_noPP/"
path_noDA = "/home/mudomini/projects/covid19.MIScnn.RESULTS/evaluation.stepwise.noDA/"
path_noPP = "/home/mudomini/projects/covid19.MIScnn.RESULTS/evaluation.stepwise.noPP/"
# Build dataset from logging data
method = ["wDwP", "wDnP", "nDwP", "nDnP"]
pathes = [path_all, path_noPP, path_noDA, path_noDAnoPP]
dt_raw = pd.DataFrame()
for i in range(0, 4):
dt_tmp = pd.DataFrame()
for fold in ["fold_0", "fold_1", "fold_2", "fold_3", "fold_4"]:
path_log = os.path.join(pathes[i], fold, "history.tsv")
fitting_log = pd.read_csv(path_log, sep="\t")
fitting_log["fold"] = fold
fitting_log["method"] = method[i]
dt_tmp = dt_tmp.append(fitting_log, ignore_index=True)
dt_raw = dt_raw.append(dt_tmp, ignore_index=True)
# Preprocessing
res_dt = pd.melt(dt_raw,
id_vars=["method", "fold", "epoch"],
value_vars=["loss", "tversky_loss", "dice_soft", "dice_crossentropy",
"val_loss", "val_tversky_loss", "val_dice_soft", "val_dice_crossentropy"],
var_name="metric", value_name="value")
# Metric selection - loss
loss_dt = res_dt[(res_dt["metric"]=="loss") | (res_dt["metric"]=="val_loss")]
# Renaming stuff
loss_dt.replace({"nDnP": "Data Augmentation: Excluded & Preprocessing: Excluded",
"wDnP": "Data Augmentation: Included & Preprocessing: Excluded",
"nDwP": "Data Augmentation: Excluded & Preprocessing: Included",
"wDwP": "Data Augmentation: Included & Preprocessing: Included"},
inplace=True)
# Plot fitting curve - v1
fig = (ggplot(loss_dt, aes("epoch", "value", color="factor(metric)"))
+ geom_smooth(method="gpr", size=1)
+ facet_wrap("method", ncol=1)
+ scale_y_continuous(limits=[0, 5])
+ ggtitle("Stepwise Fitting Curve Evaluation")
+ xlab("Number of Epochs")
+ ylab("Loss Function")
+ scale_colour_discrete(name="Dataset",
labels=["Training", "Validation"])
+ theme_bw(base_size=28))
fig.save(filename="spe.fitting_curve.png", path="evaluation",
width=20, height=16, dpi=200)
# Plot fitting curve - v2
loss_dt = res_dt[(res_dt["metric"]=="loss") | (res_dt["metric"]=="val_loss")]
loss_dt.replace({"nDnP": "Data Aug: Off & PreProc: Off",
"wDnP": "Data Aug: On & PreProc: Off",
"nDwP": "Data Aug: Off & PreProc: On",
"wDwP": "Data Aug: On & PreProc: On"},
inplace=True)
fig = (ggplot(loss_dt, aes("epoch", "value", color="factor(metric)"))
+ geom_smooth(method="gpr", size=1)
+ facet_wrap("method", ncol=2, nrow=2)
+ scale_y_continuous(limits=[0, 5])
+ ggtitle("Stepwise Fitting Curve Evaluation")
+ xlab("Number of Epochs")
+ ylab("Loss Function")
+ scale_colour_discrete(name="Dataset",
labels=["Training", "Validation"])
+ theme_bw(base_size=28))
fig.save(filename="spe.fitting_curve.alternative.png", path="evaluation",
width=16, height=12, dpi=200)
Datasets
Models
