def show_all_ipd(csvfile, mark="", alpha=1):
matrix, label, row, col = mcsv.load_matrix(csvfile)
labeltype = {
l: i
for i, l in enumerate(sorted(set([l for l in label if mark in l])))
}
x = [[] for t in labeltype]
y = [[] for t in labeltype]
if "left" == mark or "right" == mark and len(labeltype) == 2:
lvalue = list(labeltype.keys())
def filter(x, l):
xl = x[label == l]
return xl[xl != -1]
index = [
i for i, cv in enumerate(matrix.transpose())
if ttest(filter(cv, lvalue[0]), filter(cv, lvalue[1]), 10, alpha)
]
else:
index = list(range(len(col)))
for m, l, r in zip(matrix[:, index], label, row):
for mi, c in zip(m, col[index]):
if mi >= 0 and l in labeltype:
x[labeltype[l]].append(c)
y[labeltype[l]].append(mi)
for i, (xi, yi) in enumerate(zip(x, y)):
li = plt.scatter(xi, yi, color=scalarMap.to_rgba(i), s=3)
print(i, scalarMap.to_rgba(i))
plt.show()
def test_feature(csvfile, mark, ratio, alpha, count=1):
alpha = float(alpha)
ratio = float(ratio)
count = int(count)
matrix, label, row, col = mcsv.load_matrix(csvfile)
omatrix = matrix.copy()
jaccard, train_size, all_size, train_in_all = 0, [], [], 0
for i in range(count):
itrain, itest = random_split_data(ratio, matrix, label)
# matrix = normalize(matrix, label)
matrix = norm1(matrix)
labeltype = {
l: i
for i, l in enumerate(sorted(set([l for l in label if mark in l])))
}
assert "left" == mark or "right" == mark and len(labeltype) == 2
lvalue = list(labeltype.keys())
Q = alpha
itest_all = set(
np.array(range(matrix.shape[1]))[benjamini_hochberg_filter(
matrix, label == lvalue[0], label == lvalue[1], Q)])
itest_train = set(
np.array(range(
matrix[itrain, ].shape[1]))[benjamini_hochberg_filter(
matrix[itrain, ], label[itrain, ] == lvalue[0],
label[itrain, ] == lvalue[1], Q)])
#itest_all = set(ttest_matrix(matrix, label, lvalue, alpha))
#itest_train = set(ttest_matrix(matrix[itrain,], label[itrain], lvalue, alpha))
#print(itest_all)
#print(itest_train)
train_size.append(len(itest_train))
all_size.append(len(itest_all))
train_in_all += len(
itest_train.intersection(itest_all)) / len(itest_all)
jaccard += len(itest_all.intersection(itest_train)) / len(
itest_all.union(itest_train))
print(train_size)
print("Size:", all_size[0], np.mean(train_size), np.var(train_size))
print("Jaccard index:", jaccard / count, train_in_all / count)
def test_feature(csvfile, mark, alpha):
matrix, label, row, col = mcsv.load_matrix(csvfile)
omatrix = matrix.copy()
itrain, itest = random_split_data(0.8, matrix, label)
matrix = normalize(matrix, label)
labeltype = {
l: i
for i, l in enumerate(sorted(set([l for l in label if mark in l])))
}
assert "left" == mark or "right" == mark and len(labeltype) == 2
lvalue = list(labeltype.keys())
itest_all = ttest_matrix(matrix, label, lvalue, alpha)
itest_train = ttest_matrix(omatrix[itrain, ], label[itrain], lvalue, alpha)
print(itest_all)
print(itest_train)
def summary_matrix(csvfile):
from collections import defaultdict
matrix, label, row, col = mcsv.load_matrix(csvfile)
print("matrix.shape:", matrix.shape)
labelcount = defaultdict(int)
for l in label:
labelcount[l] += 1
print("Label Count:", labelcount.items())
readcount = defaultdict(int)
for r in row:
readcount[r] += 1
countread = defaultdict(list)
for r, c in readcount.items():
countread[c].append(r)
for c, r in countread.items():
print(c, len(r))
if c >= 4: print(r)