-
Notifications
You must be signed in to change notification settings - Fork 1
/
make_plots.py
147 lines (111 loc) · 4.63 KB
/
make_plots.py
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
136
137
138
139
140
141
142
143
144
145
146
import plotly as plt
import plotly.graph_objects as go
from plotly.subplots import make_subplots
import argparse
import os
import re
import math
import torch
import numpy as np
import pandas as pd
import json
import itertools
def add_trace(plot, x, y, name, color=None):
plot.add_trace(go.Scatter(x=x, y=y, name=name, line_color=color))
def add_avg_trace(plot, x, y, name, avg_epochs, color=None):
add_trace(plot, x, make_smooth(y, avg_epochs), name, color=color)
def make_smooth(y, avg_epochs=1):
ny = list()
cur_count = 0
cur_val = 0
for i in range(len(y)):
cur_count = min(cur_count + 1, avg_epochs)
if i >= avg_epochs:
cur_val -= y[i - avg_epochs]
cur_val += y[i]
ny.append(cur_val / cur_count)
return ny
def add_vertical_line(plot, x, y_st, y_en, name, color=None):
add_trace(plot, x=[x, x], y=[y_st, y_en], name=name, color=color)
def add_reward_trace(plot, plot_data, use_steps=True, avg_epochs=1, name="reward"):
if use_steps:
plot_data = plot_data[np.argsort(plot_data[:, 1])]
else:
plot_data = plot_data[np.argsort(plot_data[:, 0])]
if not len(plot_data):
return
train_episodes, steps, rewards = zip(*plot_data)
y = np.array(rewards)
if use_steps:
x = np.array(steps)
else:
x = np.array(range(1, len(train_episodes) + 1))
add_avg_trace(plot, x, y, name=name, avg_epochs=avg_epochs)
def create_reward_plot(plot_data, title="reward plot", use_steps=False, avg_epochs=1):
plot = go.Figure()
plot.update_layout(title=title)
add_reward_trace(plot, plot_data, use_steps, avg_epochs)
return plot
def load_csv(path):
dataframe = pd.read_csv(path, index_col=False)
return dataframe.to_numpy()
def get_last_log(logdir):
return max([os.path.join(logdir, d) for d in os.listdir(logdir)], key=os.path.getmtime)
def get_paths(dir, prefix=""):
return sorted(filter(lambda path: path.startswith(prefix), os.listdir(dir)),
key=lambda var:[int(x) if x.isdigit() else x for x in re.findall(r'[^0-9]|[0-9]+', var)])
def add_rewards(plot, logpath, use_steps=True, avg_constant=20, transform=lambda x: x, name="reward", env=-1):
dirpath = logpath
paths = get_paths(os.path.join(dirpath, "plots"), name)
if not paths:
paths = get_paths(os.path.join(dirpath, "plots"))
if env == -2:
for path in paths:
data = load_csv(os.path.join(logpath, "plots", path))
for env in itertools.count():
mask = (data[:, 3] == env)
datam = data[:, 0:3][mask]
if len(datam) == 0:
break
if name == "time":
datam[:, 2] -= datam[0][2]
datam[:, 2] = transform(datam[:, 2])
add_reward_trace(plot, datam, use_steps=use_steps, avg_epochs=avg_constant,
name=logpath[len("logdir"):])# + path[:-len(".csv")]) # ugly yeah
return plot
for path in paths:
data = load_csv(os.path.join(logpath, "plots", path))
if env == -1:
data = data[:, 0:3]
else:
mask = (data[:, 3] == env)
data = data[:, 0:3][mask]
if name == "time":
data[:, 2] -= data[0][2]
data[:, 2] = transform(data[:, 2])
add_reward_trace(plot, data, use_steps=use_steps, avg_epochs=avg_constant,
name=logpath[len("logdir"):])# + path[:-len(".csv")]) # ugly yeah
return plot
def create_parser():
parser = argparse.ArgumentParser()
parser.add_argument('--logpath', type=str, default=None, required=False)
parser.add_argument('--name', type=str, default="reward", required=False)
parser.add_argument('--avg', type=int, default=100, required=False)
parser.add_argument('--env', type=int, default=-1, required=False)
return parser
if __name__ == "__main__":
args = create_parser().parse_args()
if args.logpath is None:
logpaths = [get_last_log("logdir")]
else:
logpaths = list(filter(lambda x: re.fullmatch(args.logpath, x), os.listdir("logdir")))
logpaths = [os.path.join("logdir", d) for d in logpaths]
name = args.name
rewards_plot = go.Figure()
rewards_plot.update_layout(title=name, xaxis_title="Optimization step", yaxis_title=name)
use_steps = (name == "time")
transform = lambda data: data
avg_constant = args.avg
for logpath in logpaths:
add_rewards(rewards_plot, logpath, use_steps=use_steps, avg_constant=avg_constant, transform=transform, name=name, env=args.env)
plt.offline.plot(rewards_plot, filename="generated/rewards_plot.html")