def actual_ovo_classifier(classifier, train_data, test_data, output_dir,
bacteria_num, class_num):
train_answer = train_data.pop("Classification")
test_answer = test_data.pop("Classification")
train_data = train_data[general.num_to_bacteria(bacteria_num)]
test_data = test_data[general.num_to_bacteria(bacteria_num)]
classifier.fit(train_data, train_answer)
pandas.DataFrame(classifier.predict_proba(test_data),
columns=sorted(set(test_answer))).to_csv(
general.check_exist(
os.path.join(
output_dir,
"Probability_" + str(bacteria_num) + "_" +
str(class_num) + ".csv")),
index=False)
prediction = classifier.predict(test_data)
pandas.DataFrame(zip(test_answer, prediction),
columns=["real",
"prediction"]).to_csv(general.check_exist(
os.path.join(
output_dir,
"Prediction_" + str(bacteria_num) + "_" +
str(class_num) + ".csv")),
index=False)
return (bacteria_num, ) + general.aggregate_confusion_matrix(
numpy.sum(sklearn.metrics.multilabel_confusion_matrix(
test_answer, prediction),
axis=0,
dtype=int))
def headquarter_three_class_classifier(jobs=30,
input_file=None,
output_dir=None):
if (input_file is None) or (output_dir is None):
raise ValueError
elif not os.path.isfile(input_file):
raise ValueError(input_file)
data = pandas.read_csv(input_file)
data = data[["Classification"] + general.whole_values]
result_data = list()
original_class = list(data["Classification"])
for selected_class in general.three_class_combinations:
data["Classification"] = list(
map(
lambda x: "+".join(selected_class)
if x in selected_class else x, data["Classification"]))
train_data, test_data = sklearn.model_selection.train_test_split(
data,
test_size=0.1,
random_state=0,
stratify=data["Classification"])
with multiprocessing.Pool(processes=jobs) as pool:
for name, classifier in classifiers:
results = [("Number", "balanced_accuracy_score") +
general.aggregate_confusion_matrix(None)]
results += pool.starmap(
actual_three_class_classifier,
[(classifier, train_data.copy(), test_data.copy(),
os.path.join(output_dir, name), i,
general.class_to_num(selected_class))
for i in range(1, 2**len(general.absolute_values))])
results += pool.starmap(
actual_three_class_classifier,
[(classifier, train_data.copy(), test_data.copy(),
os.path.join(output_dir, name), i *
(2**len(general.absolute_values)),
general.class_to_num(selected_class))
for i in range(1, 2**len(general.relative_values))])
results = pandas.DataFrame(results[1:], columns=results[0])
results["classifier"] = name
results["combined_class"] = "-vs-".join(
sorted(set(data["Classification"])))
results.to_csv(general.check_exist(
os.path.join(output_dir, name,
"-".join(selected_class) + ".csv")),
index=False)
result_data.append(results.copy())
data["Classification"] = original_class
pandas.concat(result_data, ignore_index=True).to_csv(general.check_exist(
os.path.join(output_dir, "statistics.csv")),
index=False)
def headquarter_regressor(input_file, output_dir, watch, jobs=30):
data = pandas.read_csv(input_file)
data = data[[watch] + general.whole_values]
data.rename(columns={watch: "answer"}, inplace=True)
train_data, test_data = sklearn.model_selection.train_test_split(
data, test_size=0.1, random_state=0)
with multiprocessing.Pool(processes=jobs) as pool:
for name, regressor in regressors:
results = [("Number", "R2_score")]
results += pool.starmap(
actual_regressor,
[(regressor, train_data.copy(), test_data.copy(),
os.path.join(output_dir, name), i)
for i in range(1, 2**len(general.absolute_values))])
results += pool.starmap(
actual_regressor,
[(regressor, train_data.copy(), test_data.copy(),
os.path.join(output_dir, name), i *
(2**len(general.absolute_values)))
for i in range(1, 2**len(general.relative_values))])
results = pandas.DataFrame(results[1:], columns=results[0])
results["regressor"] = name
results["feature_num"] = list(
map(lambda x: len(general.num_to_bacteria(x)),
results["Number"]))
results.to_csv(general.check_exist(
os.path.join(output_dir, name, "statistics.csv")),
index=False)
drawing_data = pandas.concat([
pandas.read_csv(os.path.join(output_dir, name, "statistics.csv"))
for name, regressor in regressors
],
ignore_index=True)
drawing_data.to_csv(general.check_exist(
os.path.join(output_dir, "statistics.csv")),
index=False)
seaborn.set(context="poster", style="whitegrid")
fig, ax = matplotlib.pyplot.subplots(figsize=(24, 24))
seaborn.lineplot(data=drawing_data,
x="feature_num",
y="R2_score",
hue="regressor",
ax=ax,
legend="full",
hue_order=sorted(set(drawing_data["regressor"])))
fig.savefig(
general.check_exist(
os.path.join(output_dir, "Regressor_" + watch + ".png")))
matplotlib.pyplot.close(fig)
def headquarter_ovo_classifier(input_file, output_dir, jobs):
if not os.path.isfile(input_file):
raise ValueError(input_file)
elif jobs < 1:
raise ValueError(jobs)
data = pandas.read_csv(input_file)
data = data[["Classification"] + general.whole_values]
result_data = list()
for selected_class in general.two_class_combinations:
tmp_data = data.loc[(data["Classification"].isin(selected_class))]
train_data, test_data = sklearn.model_selection.train_test_split(
tmp_data,
test_size=0.1,
random_state=0,
stratify=tmp_data["Classification"])
with multiprocessing.Pool(processes=jobs) as pool:
for name, classifier in classifiers:
results = [
("Number", ) + general.aggregate_confusion_matrix(None)
]
results += pool.starmap(
actual_ovo_classifier,
[(classifier, train_data.copy(), test_data.copy(),
os.path.join(output_dir, name), i,
general.class_to_num(selected_class))
for i in range(1, 2**len(general.absolute_values))])
results += pool.starmap(
actual_ovo_classifier,
[(classifier, train_data.copy(), test_data.copy(),
os.path.join(output_dir, name), i *
(2**len(general.absolute_values)),
general.class_to_num(selected_class))
for i in range(1, 2**len(general.relative_values))])
results = pandas.DataFrame(results[1:], columns=results[0])
results["classifier"] = name
results["combined_class"] = "-vs-".join(
sorted(set(tmp_data["Classification"])))
results.to_csv(general.check_exist(
os.path.join(output_dir, name,
"-".join(selected_class) + ".csv")),
index=False)
result_data.append(results)
pandas.concat(result_data, ignore_index=True).to_csv(general.check_exist(
os.path.join(output_dir, "statistics.csv")),
index=False)
def get_tsne(csv_file=None, tsne_file=None, random_state=0):
if tsne_file is None:
tsne_file = os.path.join(default_tsne_directory,
"tsne_" + str(random_state) + ".csv")
if csv_file is None:
raise ValueError
elif not os.path.isfile(csv_file):
raise ValueError
data = pandas.read_csv(csv_file)
tsne_data = pandas.DataFrame(
sklearn.manifold.TSNE(n_components=2,
random_state=random_state,
init="pca").fit_transform(data[list(
filter(lambda x: x in tsne_columns,
list(data.columns)))]),
columns=["TSNE1", "TSNE2"])
for column in list(tsne_data.columns):
tsne_data[column] = scipy.stats.zscore(tsne_data[column])
for column in ["ID", "Classification"]:
tsne_data[column] = data[column]
tsne_data.to_csv(general.check_exist(tsne_file), index=False)
return tsne_file
def draw_violin(input_file, output_file, watch):
data = pandas.read_csv(input_file)
seaborn.set(context="poster", style="whitegrid")
fig, ax = matplotlib.pyplot.subplots(figsize=(24, 24))
seaborn.violinplot(data=data,
x="Classification",
y=watch,
order=general.classes)
statannot.add_stat_annotation(ax,
data=data,
x="Classification",
y=watch,
box_pairs=[
(general.classes[i - 1],
general.classes[i])
for i in range(1, len(general.classes))
],
test="t-test_ind",
text_format="star",
verbose=0,
order=general.classes)
fig.savefig(general.check_exist(output_file))
matplotlib.pyplot.close(fig)
def draw_extreme(csv_file, output_dir):
if not os.path.isfile(csv_file):
raise ValueError(csv_file)
statistics_data = pandas.read_csv(csv_file)
results = [("combined_class", "classifier", "bacteria", "statistics", "type", "value")]
for combined_class in sorted(set(statistics_data["combined_class"])):
tmp = list(filter(lambda x: "+" in x, combined_class.split("-vs-")))
if tmp:
combined_class_num = general.class_to_num(tmp[0].split("+"))
else:
combined_class_num = 0
for classifier in sorted(set(statistics_data["classifier"])):
prediction_directory = os.path.join(os.path.dirname(csv_file), classifier)
for statistics_value in general.aggregate_confusion_matrix(None):
selected_data = statistics_data.loc[(statistics_data["combined_class"] == combined_class) & (statistics_data["classifier"] == classifier)][[statistics_value, "Number"]]
minimum, maximum = selected_data.loc[selected_data.idxmin(axis="index")[statistics_value], "Number"], selected_data.loc[selected_data.idxmax(axis="index")[statistics_value], "Number"]
for name, value in zip(["minimum", "maximum"], [minimum, maximum]):
if combined_class_num:
prediction_data = pandas.read_csv(os.path.join(prediction_directory, "Prediction_%s_%d.csv" % (value, combined_class_num)))
else:
prediction_data = pandas.read_csv(os.path.join(prediction_directory, "Prediction_%s.csv" % (value)))
prediction_data = prediction_data.groupby(list(prediction_data.columns), as_index=False).size().reset_index().rename(columns={0: "counts"}).pivot("prediction", "real", "counts").fillna(0)
seaborn.set(context="poster", style="whitegrid")
fig, ax = matplotlib.pyplot.subplots(figsize=(24, 24))
seaborn.heatmap(prediction_data, annot=True, ax=ax, robust=True)
ax.set_title(combined_class.replace("-", " ") + " with " + statistics_value)
fig.savefig(general.check_exist(os.path.join(output_dir, name + "_" + combined_class + "_" + classifier + "_" + statistics_value + ".png")))
matplotlib.pyplot.close(fig)
results.append((combined_class, classifier, "+".join(general.num_to_bacteria(value)), statistics_value, name, value))
pandas.DataFrame(results[1:], columns=results[0]).to_csv(general.check_exist(os.path.join(output_dir, "Min_Max.csv")), index=False)
def draw_statistics(csv_file, output_dir):
if not os.path.isfile(csv_file):
raise ValueError(csv_file)
statistics_data = pandas.read_csv(csv_file)
statistics_data["feature_num"] = list(map(lambda x: len(general.num_to_bacteria(x)), statistics_data["Number"]))
for combined_class in sorted(set(statistics_data["combined_class"])):
selected_data = statistics_data.loc[(statistics_data["combined_class"] == combined_class)]
for statistics_value in sorted(general.aggregate_confusion_matrix(None)):
seaborn.set(context="poster", style="whitegrid")
fig, ax = matplotlib.pyplot.subplots(figsize=(24, 24))
seaborn.lineplot(x="feature_num", y=statistics_value, hue="classifier", ax=ax, legend="full", data=selected_data, hue_order=sorted(set(statistics_data["classifier"])), estimator="median", ci="sd")
ax.set_title(combined_class.replace("-", " "))
fig.savefig(general.check_exist(os.path.join(output_dir, "Median_" + combined_class + "_" + statistics_value + ".png")))
matplotlib.pyplot.close(fig)
seaborn.set(context="poster", style="whitegrid")
fig, ax = matplotlib.pyplot.subplots(figsize=(24, 24))
seaborn.lineplot(x="feature_num", y=statistics_value, hue="classifier", ax=ax, legend="full", data=selected_data, hue_order=sorted(set(statistics_data["classifier"])))
ax.set_title(combined_class.replace("-", " "))
fig.savefig(general.check_exist(os.path.join(output_dir, "Mean_" + combined_class + "_" + statistics_value + ".png")))
matplotlib.pyplot.close(fig)
seaborn.set(context="poster", style="whitegrid")
fig, ax = matplotlib.pyplot.subplots(figsize=(24, 24))
seaborn.lineplot(x="feature_num", y=statistics_value, hue="classifier", ax=ax, legend="full", data=selected_data, hue_order=sorted(set(statistics_data["classifier"])), estimator=min, ci=None)
ax.set_title(combined_class.replace("-", " "))
fig.savefig(general.check_exist(os.path.join(output_dir, "Min_" + combined_class + "_" + statistics_value + ".png")))
matplotlib.pyplot.close(fig)
seaborn.set(context="poster", style="whitegrid")
fig, ax = matplotlib.pyplot.subplots(figsize=(24, 24))
seaborn.lineplot(x="feature_num", y=statistics_value, hue="classifier", ax=ax, legend="full", data=selected_data, hue_order=sorted(set(statistics_data["classifier"])), estimator=max, ci=None)
ax.set_title(combined_class.replace("-", " "))
fig.savefig(general.check_exist(os.path.join(output_dir, "Max_" + combined_class + "_" + statistics_value + ".png")))
matplotlib.pyplot.close(fig)
def draw_scatter(input_file, output_file):
seaborn.set(context="poster", style="whitegrid")
fig, ax = matplotlib.pyplot.subplots(figsize=(24, 24))
seaborn.scatterplot(data=pandas.read_csv(input_file),
x="AL",
y="PD",
hue="Classification",
style="Classification",
legend="full",
ax=ax,
hue_order=general.classes)
fig.savefig(general.check_exist(output_file))
matplotlib.pyplot.close(fig)
def draw_tsne(tsne_file=None, png_file=None):
if png_file is None:
png_file = os.path.join(default_tsne_directory, "tsne.png")
if tsne_file is None:
raise ValueError(tsne_file)
elif not os.path.isfile(tsne_file):
raise ValueError(tsne_file)
seaborn.set(context="poster", style="whitegrid")
fig, ax = matplotlib.pyplot.subplots(figsize=(24, 24))
seaborn.scatterplot(data=pandas.read_csv(tsne_file),
x="TSNE1",
y="TSNE2",
hue="Classification",
style="Classification",
legend="full",
ax=ax)
fig.savefig(general.check_exist(png_file))
matplotlib.pyplot.close(fig)
