def run(self,
bins=50,
xmin=0,
xmax=30000,
step=1000,
burn=1000,
alpha=1,
output_filename=None):
# compute histogram of the input reads once for all to be used
# in the target_distribution method
self.bins = bins
self.Y, self.X = np.histogram(self.bam.df.read_length,
bins=bins,
normed=True)
lengths = self.bam_simul.df.read_length.values
self.tokeep = []
vec = []
x = self.bam.df.read_length.mean()
for i in range(self.bam_simul.df.shape[0]):
can = lengths[i]
aprob = min([
alpha,
self.target_distribution(can) / self.target_distribution(x)
])
#acceptance probability
u = pylab.uniform(0, 1)
if u < aprob:
x = can
vec.append(x)
self.tokeep.append(True)
else:
self.tokeep.append(False)
#plotting the results:
#theoretical curve
x = pylab.arange(xmin, xmax, step)
y = self.target_distribution(x)
pylab.subplot(211)
pylab.title('Metropolis-Hastings')
pylab.plot(vec)
pylab.subplot(212)
pylab.hist(vec[burn:], bins=bins, normed=1)
pylab.plot(x, y, 'r-')
pylab.ylabel('Frequency')
pylab.xlabel('x')
pylab.legend(('PDF', 'Samples'))
if output_filename is not None:
self.bam_simul.filter_bool(output_filename, self.tokeep)
def run(self, bins=50, xmin=0, xmax=30000, step=1000, burn=1000,alpha=1,output_filename=None):
# compute histogram of the input reads once for all to be used
# in the target_distribution method
self.bins = bins
self.Y, self.X = np.histogram(self.bam.df.read_length, bins=bins, density=True)
lengths = self.bam_simul.df.read_length.values
self.tokeep = []
vec = []
x = self.bam.df.read_length.mean()
for i in range(self.bam_simul.df.shape[0]):
can = lengths[i]
aprob = min([alpha,self.target_distribution(can)/self.target_distribution(x)])
#acceptance probability
u = pylab.uniform(0,1)
if u < aprob:
x = can
vec.append(x)
self.tokeep.append(True)
else:
self.tokeep.append(False)
#plotting the results:
#theoretical curve
x = pylab.arange(xmin, xmax, step)
y = self.target_distribution(x)
pylab.subplot(211)
pylab.title('Metropolis-Hastings')
pylab.plot(vec)
pylab.subplot(212)
pylab.hist(vec[burn:], bins=bins, density=1)
pylab.plot(x,y,'r-')
pylab.ylabel('Frequency')
pylab.xlabel('x')
pylab.legend(('PDF','Samples'))
if output_filename is not None:
self.bam_simul.filter_bool(output_filename, self.tokeep)
def plot_pca_vs_max_features(self, step=100, n_components=2,
progress=True):
"""
.. plot::
:include-source:
from sequana.viz.pca import PCA
from sequana import sequana_data
import pandas as pd
data = sequana_data("test_pca.csv")
df = pd.read_csv(data)
df = df.set_index("Id")
p = PCA(df)
p.plot_pca_vs_max_features()
"""
assert n_components in [2,3,4]
N = len(self.df)
if step > N:
step = N
# We start with at least 5 features
X = range(10, N, step)
from easydev import Progress
pb = Progress(len(X))
Y = []
for i, x in enumerate(X):
res = self.plot(n_components=n_components, max_features=x, show_plot=False)
Y.append(res)
if progress: pb.animate(i+1)
sub = n_components
pylab.subplot(sub,1,1)
pylab.plot(X, [y[0]*100 for y in Y])
pylab.ylabel("PC1 (%)")
pylab.subplot(sub,1,2)
pylab.plot(X, [y[1]*100 for y in Y])
pylab.ylabel("PC2 (%)")
if sub >= 3:
pylab.subplot(sub,1,3)
pylab.plot(X, [y[2]*100 for y in Y])
pylab.ylabel("PC3 (%)")
if sub >= 4:
pylab.subplot(sub,1,4)
pylab.plot(X, [y[3]*100 for y in Y])
pylab.ylabel("PC4 (%)")
def diagnostics(self, bins=60, clear=True):
if clear: pylab.clf()
pylab.subplot(3,1,1)
pylab.hist(self.aprob, bins=bins)
pylab.title("Acceptation")
pylab.subplot(3,1,2)
pylab.plot(self.vec)
pylab.title("proposition")
pylab.subplot(3,1,3)
y, x, _ = pylab.hist(self.vec, bins, density=True, lw=0.5, ec="k")
M1 = max(y)
# this normalisation is an approximation/hack
pylab.plot(self.Xtarget, self.Ytarget/ (max(self.Ytarget)/M1), "-ro")
pylab.title("simulated (blue) and target (red) distributions")