def plot_predictions(eval_labels, predictions, norm_flag, correlations, mses,
eval_names, output_directory, num_classes, cell_names,
file_txt):
for idx in range(len(eval_labels)):
y_samples_eval = eval_labels[idx]
predicted_classes = predictions[idx]
sample_names = eval_names[idx]
x_axis = np.arange(num_classes)
xticks = cell_names
plt.figure()
width = 0.35
plt.figure(figsize=(16, 9))
# plt.subplot(mum_plt_row, mum_plt_col, i + 1)
plt.bar(x_axis,
y_samples_eval,
width,
color='#f99fa1',
edgecolor='none',
label='true activity')
if norm_flag:
plt.bar(x_axis + width,
utils.minmax_scale(predicted_classes, y_samples_eval),
width,
color='#014ead',
edgecolor='none',
label='prediction')
ylabel_text = 'Normalized activity'
else:
plt.bar(x_axis + width,
predicted_classes,
width,
color='#014ead',
edgecolor='none',
label='prediction')
ylabel_text = 'Activity'
plt.xticks(x_axis + width, xticks, rotation='vertical', fontsize=9)
plt.title('{0}, correlation = {1:.4f}, mse = {2:.4f}'.format(
sample_names, correlations[idx], mses[idx]))
plt.xlabel('Cell type', fontsize=12)
plt.ylabel(ylabel_text, fontsize=12)
plt.legend()
fig = plt.gcf()
fig.tight_layout()
plt.savefig(output_directory + sample_names + file_txt + '.svg',
bbox_inches='tight')
plt.close()
def plot_predictions_subplot(eval_labels, predictions, correlations, mses,
eval_names, output_directory, num_classes,
cell_names, file_txt):
for idx in range(len(eval_labels)):
y_samples_eval = eval_labels[idx]
predicted_classes = predictions[idx]
sample_names = eval_names[idx]
# Plot
x_axis = np.arange(num_classes)
xticks = cell_names
plt.figure()
width = 0.5
fig, axs = plt.subplots(2, figsize=(24, 12))
# fig.suptitle('{0}, correlation = {1:.3f}'.format(sample_names, correlations[idx]), fontsize=18)
line_labels = ['true activity', 'prediction']
l0 = axs[0].bar(
x_axis, y_samples_eval, width,
color='#f99fa1') #, edgecolor='none', label='true activity'
l1 = axs[1].bar(
x_axis,
utils.minmax_scale(predicted_classes, y_samples_eval),
width,
color='#014ead') #, edgecolor='none', label='prediction'
axs[0].set_title('{0}, correlation = {1:.4f}, mse = {2:.4f}'.format(
sample_names, correlations[idx], mses[idx]),
fontsize=30)
fig.legend(
[l0, l1], # The line objects
labels=line_labels, # The labels for each line
loc="upper right", # Position of legend
borderaxespad=0.1, # Small spacing around legend box
fontsize=12 # Title for the legend
)
plt.xticks(x_axis, xticks, rotation='vertical', fontsize=15)
plt.xlabel('Cell type', fontsize=24)
plt.ylabel('Normalized activity', fontsize=24)
fig = plt.gcf()
fig.tight_layout()
plt.savefig(output_directory + sample_names + file_txt +
'_subplot.svg',
bbox_inches='tight')
plt.close()
def test_minmax_scaler(self):
self.assertEqual(minmax_scale([1, 1, 1]), [0, 0, 0], 'not zero list')
self.assertEqual(minmax_scale([0, 0, 0]), [0, 0, 0], 'not zero list')
self.assertEqual(
minmax_scale([1, 2, 3]).tolist(), [0., 0.5, 1.], 'wrong calculate')
self.assertEqual(
minmax_scale([0, 2, 2, 10]).tolist(), [0., 0.2, 0.2, 1.],
'wrong calculate')
self.assertIsInstance(minmax_scale([1, 2, 3]), np.ndarray,
'result is not np.ndarray')
self.assertGreaterEqual(np.min(minmax_scale([1, 20, 300])), 0,
'min value less than zero')
self.assertLessEqual(np.min(minmax_scale([1, 20, 300])), 1,
'max value more than 1')
def update_output(
rows,
timestamp_load, timestamp_col, timestamp_row, timestamp_lab, timestanp_rng, timestamp_minmax, timestamp_normal,
column_name, table_data, existing_columns, lab_column, rng_column, amount, minmax_column, a, b, normal_column
):
if rows is None:
return [{}], []
df = pd.DataFrame(rows)
time_load = string_to_datetime(timestamp_load)
time_col = string_to_datetime(timestamp_col)
time_row = string_to_datetime(timestamp_row)
time_lab = string_to_datetime(timestamp_lab)
time_rng = string_to_datetime(timestanp_rng)
time_minmax = string_to_datetime(timestamp_minmax)
time_normal = string_to_datetime(timestamp_normal)
all_times = [time_load, time_col, time_row, time_lab, time_rng, time_minmax, time_normal]
if time_load == max(all_times):
columns = [
{'id': column, 'name': column, 'deletable': True, 'renamable': True} for i, column in enumerate(df.columns)
]
return df.to_dict('records'), columns
elif time_col == max(all_times):
if column_name:
existing_columns.append({'id': column_name, 'name': column_name, 'deletable': True, 'renamable': True})
return table_data, existing_columns
elif time_row == max(all_times):
table_data.append({c['id']: '' for c in existing_columns})
return table_data, existing_columns
elif time_lab == max(all_times):
if lab_column:
df = pd.DataFrame(table_data)
df, new_col = text_to_labels(df, lab_column)
table_data = df.to_dict('records')
if new_col:
existing_columns.append({'id': new_col, 'name': new_col, 'deletable': True, 'renamable': True})
return table_data, existing_columns
elif time_rng == max(all_times):
if rng_column and amount and amount.isdigit():
df = pd.DataFrame(table_data)
df, new_col = make_ranges(df, rng_column, int(amount))
table_data = df.to_dict('records')
if new_col:
existing_columns.append({'id': new_col, 'name': new_col, 'deletable': True, 'renamable': True})
return table_data, existing_columns
elif time_minmax == max(all_times):
if minmax_column and a and b and a.isdigit() and b.isdigit():
df = pd.DataFrame(table_data)
df, new_col = minmax_scale(df, minmax_column, int(a), int(b))
table_data = df.to_dict('records')
if new_col:
existing_columns.append({'id': new_col, 'name': new_col, 'deletable': True, 'renamable': True})
return table_data, existing_columns
elif time_normal == max(all_times):
if normal_column:
df = pd.DataFrame(table_data)
df, new_col = normalize(df, normal_column)
table_data = df.to_dict('records')
if new_col:
existing_columns.append({'id': new_col, 'name': new_col, 'deletable': True, 'renamable': True})
return table_data, existing_columns
else:
return [{}], []
def plot_random_predictions(eval_labels,
predictions,
correlations,
ind_collection,
eval_names,
output_directory,
num_classes,
cell_names,
title=None,
scale=True):
for n in range(3):
mum_plt_row = 1
mum_plt_col = 1
num_plt = mum_plt_row * mum_plt_col
# 3 plots for each correlation categories
for k in range(len(ind_collection) - 1):
if len(ind_collection[k + 1]) < num_plt: continue
idx = random.sample(ind_collection[k + 1], num_plt)
y_samples_eval = eval_labels[idx]
predicted_classes = predictions[idx]
sample_names = eval_names[idx]
# Plot
x_axis = np.arange(num_classes)
if cell_names == []:
xticks = []
else:
xticks = cell_names
plt.figure(1)
width = 0.35
for i in range(num_plt):
plt.figure(figsize=(16, 9))
plt.subplot(mum_plt_row, mum_plt_col, i + 1)
plt.bar(x_axis,
y_samples_eval[i],
width,
color='#f99fa1',
edgecolor='none',
label='true activity')
if scale:
plt.bar(x_axis + width,
utils.minmax_scale(predicted_classes[i],
y_samples_eval[i]),
width,
color='#014ead',
edgecolor='none',
label='prediction')
scale_txt = '_normalized'
else:
plt.bar(x_axis + width,
predicted_classes[i],
width,
color='#014ead',
edgecolor='none',
label='prediction')
scale_txt = '_original'
plt.xticks(x_axis + width,
xticks,
rotation='vertical',
fontsize=9)
plt.title('{0}, correlation = {1:.3f}'.format(
sample_names[i], correlations[idx[i]]))
plt.xlabel('cell type', fontsize=12)
plt.ylabel(scale_txt + ' activity', fontsize=12)
plt.legend()
fig = plt.gcf()
fig.tight_layout()
if not title:
title = ''
plt.savefig(output_directory + title +
"basset_cor_q{0}{1}".format(k, n + 1) + scale_txt +
".svg",
bbox_inches='tight')
plt.close()
T -= 0.01
elif command == 'scale_true':
SCALE = True
elif command == 'scale_false':
SCALE = False
# Что показываем
for i, key in enumerate(get_best_banner(banner, top=TOP, t=T,
scale=SCALE)):
banner[key][0] += 1
print(i, key)
# Update stats
ctr = get_stats(banner)
if SCALE:
prob = softmax(minmax_scale(ctr), t=T)
else:
prob = softmax(ctr, t=T)
for i, key in enumerate(banner):
banner[key][2] = round(ctr[i], 6)
banner[key][3] = round(prob[i], 6)
command = input('Your choice: ')
# вывод статистики
if command == 'stat':
print('\n', '=' * 50, f'Temperature is = {T}', sep='\n')
print('Scale is =', SCALE, end='\n' * 2)
print("{:<6} {:<6} {:<6} {:<10} {:<10}".format('Key', 'view', 'click',
'ctr', 'prob'))