regression.py
5.22 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
#!/usr/bin/python3
# This file is supposed to propose regression models on the computation time and mem usage of the re-alignment jobs.
# Light jobs are monitored by the Monitor class in RNAnet.py, and the measures are saved in jobstats.csv.
# This was done to guess the amount of memory required to re-align the large ribosomal subunit families RF02541 and RF02543.
# INFO: Our home hardware was a 24-core VM with 50GB RAM + 8GB Swap.
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
import scipy
from sklearn.linear_model import LinearRegression
from mpl_toolkits.mplot3d import Axes3D
jobstats = pd.read_csv("data/jobstats.csv", sep=",")
families = pd.read_csv("data/statistics.csv", sep=",")
computed_families = []
comptimes = []
maxmem = []
nchains = []
maxlengths = []
for index, fam in jobstats.iterrows():
if fam["max_mem"] != -1 and fam["comp_time"] != -1:
rfam_acc = fam["label"].split(' ')[1]
computed_families.append(rfam_acc)
comptimes.append(fam["comp_time"])
maxmem.append(fam["max_mem"])
nchains.append(
families.loc[families["rfam_acc"] == rfam_acc, "total_seqs"].values[0])
maxlengths.append(
families.loc[families["rfam_acc"] == rfam_acc, "maxlength"].values[0])
nchains = [x/1000 for x in nchains] # compte en milliers de séquences
comptimes = [x/3600 for x in comptimes] # compte en heures
maxlengths = [x/1000 for x in maxlengths] # compte en kB
maxmem = [x/1024/1024 for x in maxmem] # compte en MB
summary = pd.DataFrame({"family": computed_families, "n_chains": nchains,
"max_length": maxlengths, "comp_time": comptimes, "max_mem": maxmem})
summary.sort_values("max_length", inplace=True)
summary.to_csv("summary.csv")
# ========================================================
# Plot the data
# ========================================================
fig = plt.figure(dpi=100)
plt.subplot(231)
plt.scatter(summary.n_chains, summary.max_mem)
plt.xlabel("Number of sequences (x1000 seqs)")
plt.ylabel("Peak memory (MB)")
plt.subplot(232)
plt.scatter(summary.max_length, summary.max_mem)
plt.xlabel("Maximum length of sequences (kB)")
plt.ylabel("Peak memory (MB)")
ax = fig.add_subplot(233, projection='3d')
ax.scatter(summary.n_chains, summary.max_length, summary.max_mem)
ax.set_xlabel("Number of sequences (x1000 seqs)")
ax.set_ylabel("Maximum length of sequences (kB)")
ax.set_zlabel("Peak memory (MB)")
plt.subplot(234)
plt.scatter(summary.n_chains, summary.comp_time)
plt.xlabel("Number of sequences (x1000 seqs)")
plt.ylabel("Computation time (h)")
plt.subplot(235)
plt.scatter(summary.max_length, summary.comp_time)
plt.xlabel("Maximum length of sequences (kB)")
plt.ylabel("Computation time (h)")
ax = fig.add_subplot(236, projection='3d')
ax.scatter(summary.n_chains, summary.max_length, summary.comp_time)
ax.set_xlabel("Number of sequences (x1000 seqs)")
ax.set_ylabel("Maximum length of sequences (kB)")
ax.set_zlabel("Computation time (h)")
plt.show()
# ========================================================
# Linear Regression of max_mem as function of max_length
# ========================================================
# With scikit-learn
model = LinearRegression(normalize=True, n_jobs=-1)
model.fit(np.array(summary.max_length).reshape(-1, 1), summary.max_mem)
b0 = model.intercept_
b1 = model.coef_[0]
print(f"peak_mem = {b0:.0f} + {b1:.0f} * max_length")
# with scipy
coeffs = scipy.optimize.curve_fit(lambda t, B0, B1: B0+np.exp(B1*t),
np.array(summary.max_length[:-3]), np.array(summary.max_mem[:-3]))[0]
print(f"peak_mem = {coeffs[0]:.0f} + e^({coeffs[1]:.0f} * max_length)")
coeffs_log = scipy.optimize.curve_fit(lambda t, B0, B1: B0+B1*np.log(t),
np.array(summary.max_length), np.array(summary.max_mem), p0=(400, 12000))[0]
print(
f"peak_mem = {coeffs_log[0]:.0f} + {coeffs_log[1]:.0f} * log(max_length)")
# Re-plot
x = np.linspace(0, 10, 1000)
plt.figure()
plt.scatter(summary.max_length, summary.max_mem)
plt.xlabel("Maximum length of sequences (kB)")
plt.ylabel("Peak memory (MB)")
plt.plot(x, b0 + b1*x, "-r", label="linear fit")
plt.plot(x, coeffs[0] + np.exp(coeffs[1]*x), "-g", label="expo fit on [:-3]")
plt.plot(x, coeffs_log[0] + coeffs_log[1]*np.log(x), "-b", label="log fit")
plt.ylim(0, 60000)
plt.legend()
plt.show()
print("Estimated mem required to compute RF02543 and its 11kB sequences:",
model.predict(np.array([11]).reshape(-1, 1)))
# ========================================================
# Linear Regression of comp_time as function of n_chains
# ========================================================
# With scikit-learn
model = LinearRegression(normalize=True, n_jobs=-1)
model.fit(np.array(summary.n_chains).reshape(-1, 1), summary.comp_time)
b0 = model.intercept_
b1 = model.coef_[0]
print(f"comp_time = {b0:.3f} + {b1:.3f} * n_chains")
print("Estimated computation time required for RF02543 and its 38k sequences:",
model.predict(np.array([38]).reshape(-1, 1)))
# Re-plot
x = np.linspace(0, 500, 1000)
plt.figure()
plt.scatter(summary.n_chains, summary.comp_time)
plt.xlabel("Number of sequences (x1000)")
plt.ylabel("Computation time (h)")
plt.plot(x, b0 + b1*x, "-r", label="linear fit")
plt.ylim(0, 10)
plt.legend()
plt.show()