Create analysis/plot_log.py
Browse files
baseline/analysis/plot_log.py
ADDED
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import matplotlib.pyplot as plt
|
2 |
+
import numpy as np
|
3 |
+
from scipy.optimize import curve_fit
|
4 |
+
|
5 |
+
def parse_file(file_path):
|
6 |
+
data = []
|
7 |
+
with open(file_path, 'r') as file:
|
8 |
+
for line in file:
|
9 |
+
parts = line.strip().split()
|
10 |
+
step = int(parts[0].split(':')[1].split('/')[0])
|
11 |
+
is_train = 'val' not in parts[1]
|
12 |
+
if is_train:
|
13 |
+
loss_key = 'train_loss'
|
14 |
+
else:
|
15 |
+
loss_key = 'val_loss'
|
16 |
+
loss = float(parts[1].split(':')[1])
|
17 |
+
step_avg = float(parts[3].split(':')[1].replace('ms', ''))
|
18 |
+
data.append({
|
19 |
+
'step': step,
|
20 |
+
'loss': loss,
|
21 |
+
'step_avg': step_avg,
|
22 |
+
'is_train': is_train
|
23 |
+
})
|
24 |
+
return data
|
25 |
+
|
26 |
+
# Usage
|
27 |
+
file_path = 'baseline_log.txt'
|
28 |
+
data = parse_file(file_path)
|
29 |
+
|
30 |
+
|
31 |
+
|
32 |
+
# Extract the steps and losses into separate lists
|
33 |
+
steps = np.array([d['step'] for d in filter(lambda item: item['is_train'],data)])
|
34 |
+
losses = np.array([d['loss'] for d in filter(lambda item: item['is_train'],data)])
|
35 |
+
|
36 |
+
# Take the logarithm of the data
|
37 |
+
log_steps = np.log10(steps)
|
38 |
+
log_losses = np.log10(losses)
|
39 |
+
|
40 |
+
# Define a linear function
|
41 |
+
def linear_func(x, a, b):
|
42 |
+
return a * x + b
|
43 |
+
|
44 |
+
# Fit the linear function to the logarithmic data
|
45 |
+
popt, pcov = curve_fit(linear_func, log_steps, log_losses)
|
46 |
+
|
47 |
+
# Create the plot
|
48 |
+
plt.loglog(steps, losses, label='Data')
|
49 |
+
|
50 |
+
# Plot the fitted line
|
51 |
+
x_fit = np.logspace(np.log10(np.min(steps)), np.log10(np.max(steps)), 100)
|
52 |
+
y_fit = 10 ** (popt[0] * np.log10(x_fit) + popt[1])
|
53 |
+
plt.loglog(x_fit, y_fit, label='Fitted line', color='red')
|
54 |
+
|
55 |
+
# Add title and labels
|
56 |
+
plt.title('Loss as a function of step')
|
57 |
+
plt.xlabel('Step')
|
58 |
+
plt.ylabel('Loss')
|
59 |
+
plt.legend()
|
60 |
+
|
61 |
+
# Print the fitted parameters
|
62 |
+
print('Fitted parameters: a = {:.2f}, b = {:.2f}'.format(popt[0], popt[1]))
|
63 |
+
|
64 |
+
# Save the plot to a file
|
65 |
+
plt.savefig('loss_plot2.png')
|
66 |
+
|
67 |
+
# Show the plot
|
68 |
+
plt.show()
|