def _plot_and_save_local_ancestry(df, kmer, image_filename, num_chromosomes, id_vars, x_axis, y_scale):
print('saving plot as: {}'.format(image_filename))
var_name='chromosome'
local_ancestry_df_long = pd.melt(df, id_vars=id_vars, var_name=var_name, value_name='estimated_ancestry')
new_names = {}
for i in range(1, num_chromosomes + 1):
new_names['test_{}'.format(i)] = 2*i - 2 * y_scale
new_names['true_{}'.format(i)] = 2*i - 1 * y_scale
for key, value in new_names.items():
local_ancestry_df_long.replace(key, value, inplace=True)
plot = ggplot.ggplot(ggplot.aes(x=x_axis, y=var_name, color='estimated_ancestry'), data=local_ancestry_df_long) \
+ ggplot.geom_point() \
+ ggplot.scale_y_continuous(labels=list(new_names.keys()), breaks=list(new_names.values())) \
+ ggplot.scale_color_manual(values=['#FF0000', '#0000FF', '#73008C']) \
+ ggplot.theme(plot_margin={'top':0.7, 'bottom':0.3}) ### TODO: this should depend on scale
plot.save(image_filename)
#total-based
dftmp = df[['n_sub']+brks[:5]].melt(id_vars=['n_sub'],value_vars=brks[:5], var_name = 'stat',value_name = 'value')
dftmp['method']=['(Total-Expected Total)/Expected Total']*dftmp['n_sub'].size
df_stacked = dftmp
#enhancement-based
dftmp = df[['n_sub']+brks[5:10]].melt(id_vars=['n_sub'],value_vars=brks[5:10], var_name = 'stat',value_name = 'value')
dftmp['method']=['(Enhanc-Expected Enhanc)/Expected Enhanc']*dftmp['n_sub'].size
df_stacked = df_stacked.append(dftmp)
#enhancements + full sample background
dftmp = df[['n_sub']+brks[10:]].melt(id_vars=['n_sub'],value_vars=brks[10:], var_name = 'stat',value_name = 'value')
dftmp['method']=['(Enhanc+Expected Backgr-Expected Total)/Expected Total']*dftmp['n_sub'].size
df_stacked = df_stacked.append(dftmp)
df_stacked['percentile']=['{0}th%'.format(a[1:3]) for a in df_stacked['stat']]
#plots
#compare all 3
plt1 = gg.ggplot(df_stacked, gg.aes(x='n_sub',y='value',color='percentile'))+gg.geom_line()+gg.xlab('N drives')+gg.ylab('Bias (%)')+gg.theme_bw()+gg.scale_color_manual(values=colors)+gg.geom_hline(y=[-25,25],linetype="dashed",color="gray")+gg.geom_vline(x=[10,15],linetype="dashed",color="gray")+gg.facet_wrap('method')+gg.ggtitle('Bias comparison {0}'.format(title))
plt1.save(filename = r'..\charts\drivebias_laqn_{0}.png'.format(species), width=None, height=None, dpi=300)
#plot total alone for presenation
plt2 = gg.ggplot(df_stacked[df_stacked['method']=='(Total-Expected Total)/Expected Total'], gg.aes(x='n_sub',y='value',color='percentile'))+gg.geom_line()+gg.xlab('N drives')+gg.ylab('Bias (%)')+gg.ylim(-100,100)+gg.scale_color_manual(values=colors)+gg.geom_hline(y=[-25,25],linetype="dashed",color="gray")+gg.geom_vline(x=[10,15],linetype="dashed",color="gray")+gg.ggtitle('Bias comparison {0}'.format(title))
t = gg.theme_bw()
t._rcParams['font.size']=16
plt2 = plt2+t
plt2.save(filename = r'..\charts\drivebias_laqn_{0}_total.png'.format(species), width=None, height=None, dpi=300)
#plot enhancement alone for presenation
plt3 = gg.ggplot(df_stacked[df_stacked['method']=='(Enhanc+Expected Backgr-Expected Total)/Expected Total'], gg.aes(x='n_sub',y='value',color='percentile'))+gg.geom_line()+gg.xlab('N drives')+gg.ylab('Bias (%)')+gg.ylim(-100,100)+gg.scale_color_manual(values=colors)+gg.geom_hline(y=[-25,25],linetype="dashed",color="gray")+gg.geom_vline(x=[10,15],linetype="dashed",color="gray")+gg.ggtitle('Bias comparison {0}'.format(title))
t = gg.theme_bw()
t._rcParams['font.size']=16
plt3 = plt3+t
plt3.save(filename = r'..\charts\drivebias_laqn_{0}_enhanc.png'.format(species), width=None, height=None, dpi=300)
def plot_embedding(show_plot=False, label='', n_clusters=5):
plot_df = df_tsne.copy(deep=True)
if label == '':
plt.figure(figsize=(12, 8))
plt.scatter(plot_df.iloc[:, 0], plot_df.iloc[:, 1])
plt.title('ICD Code Embeddings')
plt.savefig(base + '/Plots/icd_embedding.png')
if show_plot:
plt.show()
elif label == 'Category':
### Annotate by desc
plot_df = df_tsne.copy(deep=True)
plot_df = pd.merge(plot_df,
icd9_desc,
left_on='label',
right_on='icd9',
how='left')
plot_df.drop(['label', 'icd9'], axis=1, inplace=True)
plot_df.rename(columns={'range': 'label'}, inplace=True)
chart = ggplot(plot_df, aes(x='x1', y='x2',color = 'label') ) \
+ geom_point(size=70
,alpha=0.6
) \
+ ggtitle("Word2Vec Embeddings")
# + scale_color_manual(values = ["red", "blue","green", "purple","orange"
# , "#6DABE4","#65CBC9"
chart.save(base + '/Plots/icd_embedding_categories.png')
if show_plot:
print(chart)
elif label == 'Cluster':
#### Load embedding dict
with open(base + '/Processed_Data/icd_embedding_dict.pickle',
'rb') as handle:
embedding_dict = pickle.load(handle)
X = embedding_dict.get('embedding_matrix')
#### Cluster on Embedding Space
icd_clusters = KMeans(n_clusters=n_clusters, random_state=1).fit(X)
# Reorder Clusters
idx = pd.DataFrame(icd_clusters.labels_)[0].value_counts().index.values
lut = np.zeros_like(idx)
lut[idx] = np.arange(n_clusters)
plot_df['label'] = lut[icd_clusters.labels_].astype(str)
chart = ggplot(plot_df, aes(x='x1', y='x2',color = 'label') ) \
+ geom_point(size=70
,alpha=0.6
) \
+ ggtitle("Word2Vec Embeddings")\
+ scale_color_manual(values = ["red", "blue","green", "purple","orange"
# , "#6DABE4","#65CBC9"
])
chart.save(base + '/Plots/icd_clustered_embedding.png')
if show_plot:
print(chart)
elif label == 'Code_type':
### Annotate by desc
plot_df = df_tsne.copy(deep=True)
plot_df = pd.merge(plot_df,
icd9_desc,
left_on='label',
right_on='icd9',
how='left')
plot_df.drop(['label', 'icd9'], axis=1, inplace=True)
plot_df.rename(columns={'type': 'label'}, inplace=True)
chart = ggplot(plot_df, aes(x='x1', y='x2',color = 'label') ) \
+ geom_point(size=70
,alpha=0.6
) \
+ ggtitle("Word2Vec Embeddings")
# + scale_color_manual(values = ["red", "blue","green", "purple","orange"
# , "#6DABE4","#65CBC9"
chart.save(base + '/Plots/icd_embedding_code_type.png')
if show_plot:
print(chart)
def new_plot_ancestry_with_correct_results(test, true, y_scale=0.5, image_filename=None):
columns_to_ignore = ['POS', 'ID', 'REF', 'ALT', 'QUAL', 'FILTER', 'INFO', 'FORMAT'] ### we want only 'POS' and ancestry columns
ancestry_cols = list(filter(lambda x: x not in columns_to_ignore, test.columns))
merged = pd.DataFrame(test['POS'])
for col_name in ancestry_cols:
if col_name not in true:
raise KeyError('true ancestry dataframe is missing ancestry for id: {}'.format(col_name))
merged[col_name+'_test'] = test[col_name]
merged[col_name+'_true'] = true[col_name]
melted = pd.melt(merged, id_vars=['POS'], var_name='chromosome', value_name='ancestry')
# the above takes merged from something like this:
###
### columns: POS sample1_test sample1_true sample2_test sample2_true
### 111 pop1 pop1 pop2 pop1
### 124 pop1 pop1 pop2 pop1
###
# to this: (spaces between rows added for clarity)
###
### columns: POS chromosome ancestry
# 111 sample1_test pop1
# 124 sample1_test pop1
#
# 111 sample1_true pop1
# 124 sample1_true pop1
#
# 111 sample2_test pop2
# 124 sample2_test pop2
#
# 111 sample2_true pop1
# 124 sample2_true pop1
spacing = {}
for i, col_name in enumerate(ancestry_cols):
spacing[col_name+'_test'] = 2*i - 2 * y_scale
spacing[col_name+'_true'] = 2*i - 1 * y_scale
# taks above example to something like:
###
### columns: POS chromosome ancestry
# 111 0 pop1
# 124 0 pop1
#
# 111 1 pop1
# 124 1 pop1
#
# 111 2 pop2
# 124 2 pop2
#
# 111 3 pop1
# 124 3 pop1
for col_name, spacing_val in spacing.items():
melted.replace(col_name, spacing_val, inplace=True)
plot = ggplot.ggplot(ggplot.aes(x='POS', y='chromosome', color='ancestry'), data=melted) \
+ ggplot.geom_point() \
+ ggplot.scale_y_continuous(labels=list(spacing.keys()), breaks=list(spacing.values())) \
+ ggplot.scale_color_manual(values=['#FF0000', '#0000FF', '#73008C']) \
+ ggplot.theme(plot_margin={'top':0.7, 'bottom':0.3}) ### TODO: this should depend on scale
if image_filename is not None:
plot.save(image_filename)
else:
plot.show()
#split percentiles into different charts, all sites
#plt1 = gg.ggplot(df_along, gg.aes(x='n_passes',y='value',color='site_str'))+gg.geom_point()+gg.xlab('N drives')+gg.ylab('Bias (%)')+gg.theme_bw()+gg.xlim(0,100)+gg.facet_wrap('yparam',scales='free_y')
#plt1.save(filename = r'..\charts\bias_{0}.png'.format(c['name']), width=None, height=None, dpi=200)
#n_segments
plt2 = gg.ggplot(
df_a, gg.aes(x='n_passes', y='n_segments', color='site_str')
) + gg.geom_line() + gg.xlab('n, number drive periods') + gg.ylab(
'Sample size (number of drive patterns)') + gg.theme_bw() + gg.xlim(
0, 35) + gg.ylim(0, 2000)
plt2.save(filename=r'..\charts\n_segments_{0}_{1}.png'.format(
c['name'], dtstamp),
width=None,
height=None,
dpi=200)
#combine percentiles, split sites
plt3 = gg.ggplot(
df_along, gg.aes(x='n_passes', y='value', color='yparam')
) + gg.geom_line() + gg.xlab('n, number of drive periods') + gg.ylab(
'Sample error (%)') + gg.theme_bw() + gg.xlim(0, 35) + gg.ylim(
-100, 100) + gg.geom_hline(
y=25, linetype="dashed", color="gray") + gg.geom_hline(
y=-25, linetype="dashed", color="gray") + gg.geom_vline(
x=[10, 15], linetype="dashed",
color="gray") + gg.scale_color_manual(
values=colors) + gg.facet_wrap('site_str')
plt3.save(filename=r'..\charts\percentiles_{0}_{1}.png'.format(
c['name'], dtstamp),
width=None,
height=None,
dpi=200)