def test_localreg_realistic():
x = np.array([
-6.89438, 7.94300378, 5.5221823, 9.77749217, -0.35979986, 2.01456739,
4.80691814, 3.22260756, -7.12156073, -8.69959441
])
y = np.array([
-1.74962299, -8.55733072, 8.56537608, 1.79095858, 4.43380336,
-14.63365203, 5.41264117, 9.69660297, -13.85424098, 0.42264531
])
x0 = np.array([2., 3.])
# Testing all orders
assert np.allclose(localreg(x,
y,
x0,
degree=0,
kernel=epanechnikov,
width=1), [-14.63365203, 8.9780852],
rtol=1e-3)
assert np.allclose(localreg(x,
y,
x0,
degree=1,
kernel=epanechnikov,
width=1), [-14.5487543, 5.21322664],
rtol=1e-3)
assert np.allclose(localreg(x,
y,
x0,
degree=2,
kernel=epanechnikov,
width=1), [-14.4523815, 3.77134959],
rtol=1e-3)
# Testing width
assert np.allclose(localreg(x,
y,
x0,
degree=2,
kernel=epanechnikov,
width=2), [-14.80997735, 7.00785276],
rtol=1e-3)
# Testing frac
assert np.allclose(localreg(x,
y,
x0,
degree=2,
kernel=epanechnikov,
frac=0.5), [-6.21823369, 4.33953829],
rtol=1e-3)
def test_localreg_narrow_kernel(caplog):
x = np.array([0., 1., 2.])
y = np.array([0., 1., 2.])
x0 = np.array([0.5])
y0 = localreg(x, y, x0, degree=2, kernel=epanechnikov, width=0.4)
assert np.isnan(y0)[0]
assert (len(caplog.records) == 1)
def plot_nucleosome_occupancy_sample(fig_format ="png"):
mkdir(figure_dir)
input_fp = os.path.join(data_dir, "nucleosome_positioning", "GSM1194220_H1.bed")
df = pd.read_csv(input_fp, sep="\t", header=None).values
NROW = 4
NCOL = 4
fig, axs = plt.subplots(NROW, NROW, figsize=(NROW * EACH_SUB_FIG_SIZE, NROW * EACH_SUB_FIG_SIZE))
fig_fp = os.path.join(figure_dir, "nucleosome_occupancy_of_part_of_chr1.%s" % fig_format)
for rid in range(NROW * NCOL):
row = rid // NCOL
col = rid % NCOL
ax = axs[row][col]
start = 20000 + rid * 2000
end = 20000 + (rid + 1) * 2000
loci = df[start:end, 1].astype(np.float)
oc = df[start:end, 2].astype(np.float)
try:
y2 = localreg(loci, oc, degree=2, kernel=tricube, width=10)
ax.scatter(loci, oc, s=8, color="blue")
ax.plot(loci, y2, "k-", linewidth=2)
except np.linalg.LinAlgError as e:
sns.regplot(x=loci, y=oc, ax=ax, scatter_kws={'s': 8, 'color': "blue"})
ax.set_xlim(start, end)
ax.set_ylim(0, 600)
ax.set_title("nucleos occupancy of chr1: %d:%d" %(start, end) , fontsize=14)
plt.savefig(fig_fp, dpi=300, bbox_inches='tight', pad_inches=0.1)
def plot_local_regression_and_RD(fig_format="png"):
K_RD = [1] # , 0
N_COL = N_ROW = 2
cm = plt.get_cmap('gist_rainbow')
mkdir(figure_dir)
for km in K_RD:
correlation_type = '' if km else '-methy'
for rid, region_label in enumerate(REGION_LABELS):
input_dir = os.path.join(data_dir, "Q1_CGI_density", region_label)
REGIONS = ["Non-" + region_label, region_label]
fig, axs = plt.subplots(N_ROW,
N_COL,
figsize=(N_COL * EACH_SUB_FIG_SIZE,
N_ROW * EACH_SUB_FIG_SIZE))
fig_fp = os.path.join(figure_dir,
"%s.%s" % (region_label, fig_format))
for iid, region in enumerate(REGIONS):
for jid, cgi in enumerate(CGI_LABELS):
correlation_fp = os.path.join(
input_dir,
region + "-" + cgi + correlation_type + "-Rd.bed")
file_label = region + " " + cgi
idx = iid * len(REGIONS) + jid
row = iid
col = jid
ax = axs[row][col]
RD_df = pd.read_csv(correlation_fp, sep="\t",
header=None).values
x = RD_df[:, 0]
y = RD_df[:, 1]
color = cm(1. * idx / (N_COL * N_ROW))
try:
y2 = localreg(x,
y,
degree=2,
kernel=tricube,
width=100)
ax.scatter(x, y, s=8, color=color, label=file_label)
ax.plot(x, y2, "k-", linewidth=2)
except np.linalg.LinAlgError as e:
sns.regplot(x=x,
y=y,
ax=ax,
scatter_kws={
's': 8,
'color': color
}) #, line_kws ={'color':'black', "lw": 2}
ax.set_xticks(range(0, D_MAX + 1, 200))
ax.set_xlim(0, D_MAX)
ax.set_ylim(0, 1.0)
ax.set_title(file_label, fontsize=14)
plt.savefig(fig_fp,
dpi=300,
bbox_inches='tight',
pad_inches=0.1)
def plot_local_regression_and_RD_within_vs_across(fig_format="png"):
K_RD = [1]# , 0
N_COL = 4
N_ROW = 2
cm = plt.get_cmap('gist_rainbow')
mkdir(figure_dir)
for km in K_RD:
correlation_type = '' if km else '-methy'
fig, axs = plt.subplots(N_ROW, N_COL, figsize=(N_COL * EACH_SUB_FIG_SIZE, N_ROW * EACH_SUB_FIG_SIZE))
fig_fp = os.path.join(figure_dir, "CGI_RATIO_Within-Across-5000bp.%s" % fig_format)
for rid, ratio in enumerate(RATIOS):
within_correlation_fp = os.path.join(data_dir, "CGI_identified_with_different_thereshold", "CGI_%s_K_intersected" % ratio + correlation_type +"-only-within-Rd.bed")
across_correlation_fp = os.path.join(data_dir, "CGI_identified_with_different_thereshold", "CGI_%s_K_intersected" % ratio + correlation_type +"-both-within-across-Rd.bed")
file_label = "Obs/Expect " + str(RATIO_LABELS[rid])
row = rid // N_COL
col = rid % N_COL
ax = axs[row][col]
within_RD = pd.read_csv(within_correlation_fp, sep="\t", header=None).values
across_RD = pd.read_csv(across_correlation_fp, sep="\t", header=None).values
x1 = within_RD[:, 0]
y1 = within_RD[:, 1]
x2 = across_RD[:, 0]
y2 = across_RD[:, 1]
color = cm(1. * rid / len(RATIOS))
try:
yy1 = localreg(x1, y1, degree=2, kernel=tricube, width=100)
yy2 = localreg(x2, y2, degree=2, kernel=tricube, width=100)
ax.scatter(x1, y1, s=3, color=color, label="within")
ax.scatter(x2, y2, s=1.5, color="black", label="within")
ax.plot(x2, yy2, "w-", linewidth=1.5)
ax.plot(x1, yy1, "k-", linewidth=1.5)
except np.linalg.LinAlgError as e:
sns.regplot(x=x1, y=y1, ax=ax, scatter_kws={'s':8, 'color': color})#, line_kws ={'color':'black', "lw": 2}
sns.regplot(x=x2, y=y2, ax=ax, scatter_kws={'s':8, 'color': "black"})#, line_kws ={'color':'black', "lw": 2}
ax.set_xticks(range(0, D_MAX + 1, 1000))
ax.set_xlim(0, D_MAX)
ax.set_ylim(0, 1.0)
ax.set_title(file_label, fontsize=14)
plt.savefig(fig_fp, dpi=300, bbox_inches='tight', pad_inches=0.1)
def plot_local_regression_and_RD(fig_format="png"):
K_RD = [1]# , 0
N_COL = 1
N_ROW = 1
plt.rc('xtick', labelsize=12) # fontsize of the tick labels
plt.rc('ytick', labelsize=12) # fontsize of the tick labels
cm = plt.get_cmap('gist_rainbow')
mkdir(figure_dir)
for km in K_RD:
correlation_type = '' if km else '-methy'
fig, ax = plt.subplots(N_ROW, N_COL, figsize=((N_COL * EACH_SUB_FIG_SIZE), N_ROW * EACH_SUB_FIG_SIZE))
fig_fp = os.path.join(figure_dir, "CGI_RATIO_FROM-06-13-%d.%s" % (D_MAX, fig_format))
for rid, ratio in enumerate(RATIOS):
correlation_fp = os.path.join(data_dir, "CGI_identified_with_different_thereshold", "CGI_%s_K_intersected" % ratio + correlation_type +"-only-within-Rd.bed")
file_label = "Obs/Exp>" + str(RATIO_LABELS[rid])
# row = rid // N_COL
# col = rid % N_COL
# if N_ROW == 1:
# ax = axs[0]
# else:
# ax = axs[row][0]
RD_df = pd.read_csv(correlation_fp, sep="\t", header=None).values
x = RD_df[:, 0]
y = RD_df[:, 1]
color = cm(1. * rid / len(RATIOS))
try:
y2 = localreg(x, y, degree=2, kernel=tricube, width=100)
ax.scatter(x, y, s=8, color=color, label=file_label)
ax.plot(x, y2, color='black', linestyle = 'solid', linewidth=2)
except np.linalg.LinAlgError as e:
sns.regplot(x=x, y=y, ax=ax, scatter_kws={'s':8, 'color': color})#, line_kws ={'color':'black', "lw": 2}
ax.set_xticks(range(0, D_MAX + 1, 200))
ax.set_xlim(0, D_MAX)
ax.set_ylim(0, 1.0)
ax.set_xlabel("Genomic Distance(bp)", fontsize=18, fontweight='bold')
ax.set_ylabel("Pearson Correlation", fontsize=18, fontweight='bold')
# box = ax.get_position()
# ax.set_position([box.x0, box.y0, box.width * 0.8, box.height])
ax.legend(loc='best')#, bbox_to_anchor=(1, 0.5))
# ax.set_title(file_label, fontsize=14)
plt.savefig(fig_fp, dpi=300)
def smooth(x, y, xgrid):
""" Approximates the linear relationship between x and y by a random subsample
:param (np.array) x: x-axis data points
:param np.array y: y-axis data points
:param np.array xgrid: interpolation points
:return np.array: approximate/smoothed y-values
"""
samples = np.random.choice(len(x), len(x), replace=True)
y_s = y[samples]
x_s = x[samples]
y_sm = localreg(x_s,
y_s,
x0=None,
degree=1,
kernel=triangular,
width=19.08094)
y_grid = scipy.interpolate.interp1d(x_s, y_sm,
fill_value='extrapolate')(xgrid)
return y_grid
def plot_whole_landscape(fig_format="png"):
regions = ["Genomic_Regions", "Histone_Modification", "ChromHMM",
"TFBS"] #
RD_DIRNAME = "Whole_Landscape"
cm = plt.get_cmap('gist_rainbow')
FIG_DIR = os.path.join(BASE_DIR, "figures")
NBIN = 30
vmins = [0, 120]
vmaxs = [0.15, 220]
N_COL = 18
COL_Labels = [
"Corr", "Corr with DNase", "Corr with Nuc Occ", "Hist of k",
"Hist of methylation", "Hist2D of k/methy", "DNase Peak",
"Nucleosome Occupancy", "H3k4me1", "H3k4me3", "H3k9me3", "H3k9ac",
"H3k27ac", "H3k27me3", "H3k36me3", "H4k20me", "CTCF", "p300"
]
COL_INDEXS = [
0, 0, 0, 4, 3, 0, -2, -1, 6, 8, 9, 10, 11, 12, 13, 14, 15, 16
]
MP = 50 #Max Peak Value
xlims = [
D_MAX, D_MAX, D_MAX, 10, 1., 1, 0.15, 220, MP, MP, MP, MP, MP, MP, MP,
MP, MP, MP
]
ylims = [
1, 1, 1, 0.6, 16, 1, 70, 0.05, 0.25, 0.2, 0.14, 0.06, .1, .12, .25,
.20, 0.05, .2
]
for REGION in regions:
fig_dir = os.path.join(FIG_DIR, RD_DIRNAME)
mkdir(fig_dir)
out_rd_corr_dir = os.path.join("../data/K_region_intersect", REGION,
"K_Rd")
file_paths = [
os.path.join(out_rd_corr_dir, "%s.bed" % file_name)
for file_name in FILE_ORDERED_NAMES[REGION]
]
file_labels = FILE_LABELS[REGION]
bed_fps = [
os.path.join("../data/K_region_intersect", REGION,
"%s.bed" % file_name)
for file_name in FILE_ORDERED_NAMES[REGION]
]
N_ROW = len(file_labels)
fig_fp = os.path.join(fig_dir, "%s.%s" % (REGION, fig_format))
fig, axs = plt.subplots(N_ROW,
N_COL,
figsize=(N_COL * EACH_SUB_FIG_SIZE,
N_ROW * EACH_SUB_FIG_SIZE))
for row in range(N_ROW):
RD_df = pd.read_csv(file_paths[row], sep="\t", header=None).values
bed_df = pd.read_csv(bed_fps[row], sep="\t", header=None).values
for col in range(N_COL):
print("%s, %s" % (file_labels[row], COL_Labels[col]))
ax = axs[row][col]
xlim = xlims[col]
ylim = ylims[col]
if col <= 2:
x = RD_df[:, 0]
y = RD_df[:, 1]
try:
y2 = localreg(x,
y,
degree=2,
kernel=tricube,
width=100)
if col != 0:
z = RD_df[:, 1 + col]
sc = ax.scatter(x,
y,
s=8,
c=z,
label=file_labels[row],
cmap=cm,
vmin=vmins[col - 1],
vmax=vmaxs[col - 1])
fig.colorbar(sc, ax=ax)
else:
ax.scatter(x,
y,
s=2,
color=cm(1. * row / N_ROW),
label=file_labels[row])
ax.plot(x, y2, "k-", linewidth=2)
except np.linalg.LinAlgError as e:
sns.regplot(
x=x,
y=y,
ax=ax,
scatter_kws={
's': 8,
'color': cm(1. * row / N_ROW)
}) # , line_kws ={'color':'black', "lw": 2}
ax.set_xticks(range(0, D_MAX + 1, 200))
ax.set_xlim(0, xlim)
ax.set_ylim(0, ylim)
ax.set_title(COL_Labels[col], fontsize=18)
if col == 0:
ax.set_ylabel(file_labels[row], fontsize=20)
if row == N_ROW - 1:
ax.set_xlabel("Genomic Distance(bp)")
else:
col_ind_in_bed = COL_INDEXS[col]
if col_ind_in_bed != 0:
vals = bed_df[:, col_ind_in_bed]
vals = vals[vals != "."].astype(float)
vals[vals > xlim] = xlim
_ = ax.hist(vals,
bins=NBIN,
density=True,
color=cm(1. * col / N_COL),
edgecolor='black',
alpha=0.5,
linewidth=0.5)
ax.set_xlim([0, xlim])
ax.set_ylim(0, ylim)
ax.set_title(COL_Labels[col], fontsize=18)
else:
ks = bed_df[:, 4].astype(float)
methys = bed_df[:, 3].astype(float)
ks[ks <= 0] = 0.001
ks = np.log10(ks)
h = ax.hist2d(methys,
ks,
bins=(NBIN, NBIN),
density=True,
vmin=0,
vmax=5,
cmap="viridis")
fig.colorbar(h[3], ax=ax)
if col == 0:
ax.set_ylabel('log10(K)')
if row == N_ROW - 1:
ax.set_xlabel('Methylation level')
ax.set_ylim(-1, 1)
ax.set_xlim(0, 1)
ax.set_xticks([0.2 * i for i in range(6)])
plt.savefig(fig_fp, dpi=300, bbox_inches='tight', pad_inches=0.1)
# p_upper[0] = 12
# p_init[0] = 9.11
# # p_lower[2] = 1
# # p_upper[2] = 2
# # p_init[2] = 1.5
# p_upper[3] = 3
# if np.abs(l-2e-3)<1e-4:
# p_lower[1] = 0
# p_upper[1] = 1e-6
# p_init[1] = 0
# p_lower[3] = -1
if inspect or fit_of_fit:
zeta_ = np.linspace(0, lambd, 5000)
g_ = localreg(zeta, g, zeta_, kernel=gaussian, width=0.1, degree=2)
bounded = ""
model = partial(additive_model, lambd)
model = lambda zeta, C, A, alpha, delta: additive_model(lambd, zeta, C, A, alpha, 0, 1, 1, delta)
if fit_of_fit:
zeta_fit = zeta_
g_fit = g_
else:
zeta_fit = zeta
g_fit = g
zeta_bnd = 5
frac_bnd = 0.5
inds = np.where(np.logical_or(zeta_fit<zeta_bnd, zeta_fit>lambd-zeta_bnd))[0]
n_bnd = len(inds)
def rdd_plot(data, sbins, bw, k, calc_points, dependant_var):
"""Plots smoothed local regression with bootstrapped CIs on both sides of "margin_1".
:param (df) data: df that contains "margin_1" and parameter dependant_var
:param (int) sbins: length of bin
:param (int) bw: bandwidth for local regression
:param (int) k: iterations of resampling by bootstrapping
:param (int) calc_points: points where to calculate smoothed value
:param (str) dependant_var: name of dependant variable in df
:return: plot
"""
temp_df = bin_fct(data, sbins)
avg_rank_impr = temp_df.groupby(temp_df["bin"]).mean()[dependant_var]
x = range(-30, 30, sbins)
vic_marg = x - np.mod(x, sbins) + sbins / 2
df_figure2 = pd.DataFrame([vic_marg, avg_rank_impr],
index=["vic_marg", "rank_imp"]).transpose()
df_neg = temp_df.loc[(temp_df["margin_1"] < 0)].sort_values(
by=["margin_1"])
df_pos = temp_df.loc[(temp_df["margin_1"] > 0)].sort_values(
by=["margin_1"])
y1 = np.asarray(df_neg[dependant_var])
y2 = np.asarray(df_pos[dependant_var])
x1 = np.asarray(df_neg["margin_1"])
x2 = np.asarray(df_pos["margin_1"])
x_sm1 = x1[0::calc_points]
x_sm2 = x2[0::calc_points]
reg_1 = localreg(x1, y1, x0=x_sm1, degree=1, kernel=triangular, width=bw)
reg_2 = localreg(x2, y2, x0=x_sm2, degree=1, kernel=triangular, width=bw)
xgrid1 = np.linspace(-30, 0, 50)
xgrid2 = np.linspace(0, 30, 50)
smooths1 = np.stack([smooth(x1, y1, xgrid1) for i in range(k)]).T
smooths2 = np.stack([smooth(x2, y2, xgrid2) for i in range(k)]).T
mean_neg = np.nanmean(smooths1, axis=1)
stderr_neg = np.nanstd(smooths1, axis=1, ddof=0)
mean_pos = np.nanmean(smooths2, axis=1)
stderr_pos = np.nanstd(smooths2, axis=1, ddof=0)
fig, (ax0) = plt.subplots(1, 1, figsize=(12, 8), tight_layout=True)
plt.fill_between(xgrid1,
mean_neg - 1.96 * stderr_neg,
mean_neg + 1.96 * stderr_neg,
alpha=0.25)
plt.fill_between(xgrid2,
mean_pos - 1.96 * stderr_pos,
mean_pos + 1.96 * stderr_pos,
alpha=0.25)
plt.axvline(0, linewidth=0.4, color='r')
ax0.grid(True)
ax0.scatter(df_figure2["vic_marg"], df_figure2["rank_imp"])
ax0.plot(x_sm1, reg_1)
ax0.plot(x_sm2, reg_2)
plt.xlabel("% Margin of Victory")
plt.ylabel("Average Rank Improvment")
ax0.axis([-30, 30, -6, 6])
plt.show()
x = np.linspace(0, L, 5000)
x0 = np.linspace(0, L, 500)
yf = np.sin(x**2)
y = yf + 0.5 * np.random.randn(*x.shape)
plt.plot(x, yf, label='$\\sin(x^2)$')
plt.plot(x, y, '+', markersize=0.2, color='black')
N = 200
yarr = np.zeros((N, len(x0)))
for n in range(N):
ind = np.random.randint(0, len(x), len(x))
xb = x[ind]
yb = y[ind]
yarr[n, :] = localreg(xb, yb, x0, degree=2, kernel=tricube, width=0.4)
# plt.plot(x0, yarr[n,:], linewidth=0.5, color='black')
lower = np.percentile(yarr, 2.5, axis=0)
upper = np.percentile(yarr, 97.5, axis=0)
plt.fill_between(x0, lower, upper, color='gray', alpha=0.5)
y0 = localreg(x, y, x0, degree=2, kernel=tricube, width=0.4)
plt.plot(x0, y0, label='Local regression')
# ym = np.average(yarr, axis=0)
# plt.plot(x0, ym, '--', label='Bootstrapped local regression')
plt.title('Locally Weighted Polynomial Regression')
plt.xlabel('x')
plt.xlabel('y')
print('linear dim:' + str(semi_data.linear_dim))
print('varying dim:' + str(semi_data.varying_dim))
I = 5
la_design = lae.la_design(semi_data.varying, I)
varying_hat, linear_hat, index_design = lae.full_estimate(semi_data, I)
print('constant coefficients:')
print(linear_hat)
"""
further estimate via local polynomial regression
"""
target = varying_hat[:, 1]
index = index_design
a_hat = localreg(index, target, degree=1, kernel=epanechnikov, width=0.3)
plt.style.use('ggplot')
plt.plot(index, a_hat, label='Local average + Local linear')
"""example for varying coefficient model"""
import numpy as np
from numpy import pi, sin, cos, exp, sqrt, std
from laepy import lae
from localreg import *
import matplotlib.pyplot as plt
n, d, snr = 500, 2, 5
x = np.random.normal(0, 1, [n, d])
u = np.random.uniform(0, 1, n)
coef = np.array([sin(60 * u), 4 * u * np.subtract(1, u)])
def plot_local_regression_and_RD_separately(max_d, fig_format="png"):
regions = ["Genomic_Regions", "Histone_Modification", "ChromHMM",
"TFBS"] #
K_RD = [3] # 0, 1, 2
N_COL = 5
cm = plt.get_cmap('gist_rainbow')
FIG_DIR = os.path.join(BASE_DIR, "figures")
plt.rc('xtick', labelsize=12) # fontsize of the tick labels
plt.rc('ytick', labelsize=12) # fontsize of the tick labels
for REGION in regions:
fig_dir = os.path.join(FIG_DIR, "Rd")
mkdir(fig_dir)
for km in K_RD:
if km == 1:
inter_name = "K_Rd"
elif km == 2:
inter_name = "f_Rd"
elif km == 3:
inter_name = "DNase_Rd"
else:
inter_name = "Methy_Rd"
out_rd_corr_dir = os.path.join("../data/K_region_intersect",
REGION, inter_name)
file_paths = [
os.path.join(out_rd_corr_dir, "%s.bed" % file_name)
for file_name in FILE_ORDERED_NAMES[REGION]
]
file_labels = FILE_LABELS[REGION]
N_FILES = len(file_labels)
N_ROW = int(math.ceil((N_FILES) / N_COL))
fig_fp = os.path.join(
fig_dir, "%s_%s.%s" % (REGION, inter_name, fig_format))
fig, axs = plt.subplots(N_ROW,
N_COL,
figsize=(N_COL * (EACH_SUB_FIG_SIZE - 1),
N_ROW * (EACH_SUB_FIG_SIZE - 1)))
for j in range(N_FILES):
row = j // N_COL
col = j % N_COL
if N_ROW == 1:
ax = axs[col]
else:
ax = axs[row][col]
file_path = file_paths[j]
RD_df = pd.read_csv(file_path, sep="\t", header=None).values
x = RD_df[:, 0]
y = RD_df[:, 1]
try:
y2 = localreg(x, y, degree=2, kernel=tricube, width=100)
if file_labels[j] == "Genome":
ax.scatter(x,
y,
s=8,
label=file_labels[j],
color="blue")
ax.plot(x, y2, "w-", linewidth=2)
else:
ax.scatter(x,
y,
s=8,
label=file_labels[j],
color=cm(1. * j / N_FILES))
ax.plot(x, y2, "k-", linewidth=2)
except np.linalg.LinAlgError as e:
sns.regplot(x=x,
y=y,
ax=ax,
scatter_kws={
's': 8,
'color': cm(1. * j / N_FILES)
}) #, line_kws ={'color':'black', "lw": 2}
ax.set_xticks(range(0, max_d + 1, 200))
ax.set_xlim(0, max_d)
ax.set_ylim(0, 1.0)
ax.set_title(file_labels[j], fontsize=18)
plt.savefig(fig_fp, dpi=300, bbox_inches='tight', pad_inches=0.1)
def plot_local_regression_and_RD(max_d, fig_format="png"):
regions = [
"Genomic_Regions", "Histone_Modification", "ChromHMM", "TFBS"
] #"Genomic_Regions", "Histone_Modification", "ChromHMM", "TFBS"
K_RD = [3] # , 0
N_COL = 5
cm = plt.get_cmap('gist_rainbow')
FIG_DIR = os.path.join(BASE_DIR, "figures")
fig_dir = os.path.join(FIG_DIR, "Rd")
mkdir(fig_dir)
for REGION in regions:
fig_fp = os.path.join(fig_dir, "%s.%s" % (REGION, fig_format))
fig, axs = None, None
for km in K_RD:
if km == 1:
inter_name = "K_Rd"
color = "r"
c = "red"
label = "k"
elif km == 2:
inter_name = "f_Rd"
color = "g"
c = "green"
label = "f"
else:
inter_name = "Methy_Rd"
color = "b"
c = "blue"
label = "methy"
out_rd_corr_dir = os.path.join("../data/K_region_intersect",
REGION, inter_name)
file_paths = [
os.path.join(out_rd_corr_dir, "%s.bed" % file_name)
for file_name in FILE_ORDERED_NAMES[REGION]
]
file_labels = FILE_LABELS[REGION]
N_FILES = len(file_labels)
N_ROW = int(math.ceil((N_FILES) / N_COL))
if not fig:
fig, axs = plt.subplots(
N_ROW,
N_COL,
figsize=(N_COL * EACH_SUB_FIG_SIZE,
N_ROW * (EACH_SUB_FIG_SIZE - 1)))
for j in range(N_FILES):
row = j // N_COL
col = j % N_COL
if N_ROW == 1:
ax = axs[col]
else:
ax = axs[row][col]
file_path = file_paths[j]
RD_df = pd.read_csv(file_path, sep="\t", header=None).values
x = RD_df[:, 0]
y = RD_df[:, 1]
try:
y2 = localreg(x, y, degree=2, kernel=tricube, width=100)
sc = ax.scatter(x, y, s=8, label=label, color=c)
ax.plot(x, y2, "w-", linewidth=2)
except np.linalg.LinAlgError as e:
sns.regplot(x=x,
y=y,
ax=ax,
scatter_kws={
's': 8,
'color': c
},
line_kws={
'color': 'white',
"lw": 2
})
ax.set_xticks(range(0, max_d + 1, 200))
ax.set_xlim(0, max_d)
ax.set_ylim(0, 1.0)
ax.set_title(file_labels[j], fontsize=16)
if km == 2 and j == 0:
ax.legend()
plt.savefig(fig_fp, dpi=300) #, bbox_inches='tight', pad_inches=0.1
import numpy as np import matplotlib.pyplot as plt from localreg import * np.random.seed(1234) x = np.linspace(1.5, 5, 2000) yf = np.sin(x * x) y = yf + 0.5 * np.random.randn(*x.shape) y0 = localreg(x, y, degree=0, kernel=rbf.tricube, width=0.3) y1 = localreg(x, y, degree=1, kernel=rbf.tricube, width=0.3) y2 = localreg(x, y, degree=2, kernel=rbf.tricube, width=0.3) plt.plot(x, y, '+', markersize=0.6, color='gray') plt.plot(x, yf, label='Ground truth ($\sin(x^2)$)') plt.plot(x, y0, label='Moving average') plt.plot(x, y1, label='Local linear regression') plt.plot(x, y2, label='Local quadratic regression') plt.legend() plt.show()