def check_replica(self):
#discard everything below the flag
new_experiment = []
new_replica = []
min_value = sys.maxint
max_value = -sys.maxint
for i in range(len(self.replica_values)):
if self.experiment_values[i] > self.count_filter and self.replica_values[i] > self.count_filter:
new_experiment.append(math.log(self.experiment_values[i], 2))
new_replica.append(math.log(self.replica_values[i], 2))
min_value = min(min_value, math.log(self.experiment_values[i], 2), math.log(self.replica_values[i], 2))
max_value = max(max_value, math.log(self.experiment_values[i], 2), math.log(self.replica_values[i], 2))
#print self.replica_values
self.experiment_values = new_experiment
self.replica_values = new_replica
try:
if self.postscript:
import matplotlib
matplotlib.use("PS")
from matplotlib.pyplot import plot, show, xlabel, ylabel, axhline, axis, clf, text, title, xlim, ylim
except:
__matplotlibwarn(self)
return 0
clf()
r_squared = utils.pearson(self.experiment_values, self.replica_values)**2
text(min_value+abs(max_value)*0.1, max_value-abs(max_value)*0.2, r'Pearson $R^2$= %s'%round(r_squared, 3), fontsize=18, bbox={'facecolor':'yellow', 'alpha':0.5, 'pad':10})
xlabel("log2(%s)"%self.experiment_label, fontsize=18)
ylabel("log2(%s)"%self.replica_label, fontsize=18)
xlim(min_value, max_value)
ylim(min_value, max_value)
title(self.title_label, fontsize=24)
plot(self.experiment_values, self.replica_values, '.')
self._save_figure("check_replica")
def evaluate(model, loader):
model.eval()
y_hat_list = []
y_list = []
for batch_data in loader:
a2a_g, b2a_g, b2b_gl, feats, types, counts, y = batch_data
_, y_hat = model(a2a_g, b2a_g, b2b_gl, types, counts)
y_hat_list += y_hat.tolist()
y_list += y.tolist()
y_hat = np.array(y_hat_list).reshape(-1,)
y = np.array(y_list).reshape(-1,)
return rmse(y, y_hat), mae(y, y_hat), sd(y, y_hat), pearson(y, y_hat)
def check_replica_correlation(self):
"No usado, de momento"
min_tags = 20
experiment_reader = utils.read_fetcher(self.current_experiment_path, self.experiment_format, cached=self.cached, logger=self.logger, use_samtools=self.use_samtools, access_sequential=self.access_sequential)
replica_reader = utils.read_fetcher(self.current_replica_path, self.experiment_format, cached=self.cached, logger=self.logger, use_samtools=self.use_samtools, access_sequential=self.access_sequential)
correlations_acum = 0
num_correlations = 0
for region_line in open(self.region_path):
sline = region_line.split()
region_experiment = self._region_from_sline(sline)
region_replica = region_experiment.copy()
tags_experiment = experiment_reader.get_overlaping_clusters(region_experiment, overlap=1)
tags_replica = replica_reader.get_overlaping_clusters(region_experiment, overlap=1)
count_experiment = len(tags_experiment)
count_replica = len(tags_replica)
correlations = []
if count_experiment+count_replica > min_tags:
region_experiment.add_tags(tags_experiment, clusterize=True)
region_replica.add_tags(tags_replica, clusterize=True)
num_correlations += 1
correlation = utils.pearson(region_experiment.get_array(), region_replica.get_array())
correlations_acum += max(0, correlation)
correlations.append(correlation)
print correlations_acum/num_correlations
try:
if self.postscript:
import matplotlib
matplotlib.use("PS")
from matplotlib.pyplot import plot, boxplot, show, legend, figure, xlabel, ylabel, subplot, axhline, axis
except:
__matplotlibwarn(self)
return 0
print correlations
boxplot(correlations)
self._save_figure("check_replica")
import numpy as np
from datasets import read_arcene
from sklearn.preprocessing import LabelBinarizer
import matplotlib.pyplot as plt
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import accuracy_score, f1_score
from sklearn.svm import SVC
from svm import SVM
from optim import GD
from utils import (pearson, spearman, chi2)
metrics = {
'pearson': pearson,
'spearman': spearman,
'chi2': chi2,
'together': lambda x, y: pearson(x, y) * spearman(x, y) * chi2(x, y)
}
def sort_features(met, n):
features = []
inds = np.arange(n)
for m in met:
crit = metrics[m]
features.append(
np.array(sorted(inds, key=lambda i: crit(X[:, i], y),
reverse=True)))
return features
