Log curves and plots¶

Comet allows you to log curves and plots to the Comet platform for visual analysis of an experiment's attributes and performance.

You can use log_curve() to log curves as x and y vectors that can be plotted against one another, or directly log plots created with Seaborn, Plotly and Matplotlib figures using the log_figure() method. You can visualize logged curves and plots in the Single Experiment page and in panels.

The following methods can be used to log plots and curves:

• For plots: log_figure().
• For curves: log_curve().

The following panels can be used to visualize plots and curves:

• For curves: Logged Curve.
• For plots: Image Panel.

In addition you can view plots and curves in the following Single Experiment page tabs:

• For the first 25 curves: Panels Tab.
• For plots: Graphics Tab.
• For curves and plots: Assets & Artifacts Tab: curves and images folder respectively.

Log a curve¶

The example below logs a precision-recall curve from precision and recall metrics calculated with the sklearn.metrics.precision_recall_curve() method on 4 mock true and predicted labels.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20

import comet_ml
import numpy as np
from sklearn.metrics import precision_recall_curve

# Initialize the Comet Experiment
comet_ml.login()
exp = comet_ml.start()

# Create an example x and y vectors for the curve
y_true = np.array([0, 0, 1, 1])
y_predicted = np.array([0.1, 0.4, 0.35, 0.8])
precision, recall, _ = precision_recall_curve(y_true, y_predicted)

# Log the curve to Comet
exp.log_curve(
    name="precision-recall",
    x=precision,
    y=recall,
    step=1,
)

Additionally, you could use the overwrite argument to overwrite any existing asset with the same name.

Log a figure¶

The examples below showcases how to log example figures created with Matplotlib, Plotly, and Seaborn.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22

import comet_ml
import numpy as np
import matplotlib.pyplot as plt

# Initialize the Comet Experiment
comet_ml.login()
exp = comet_ml.start()

# Create an example scatter plot figure
data = {
    'a': np.arange(50),
    'c': np.random.randint(0, 50, 50),
    'd': np.random.randn(50)
}
data['b'] = data['a'] + 10 * np.random.randn(50)
data['d'] = np.abs(data['d']) * 100
fig = plt.figure(figsize=(5, 5))
plt.scatter('a', 'b', c='c', s='d', data=data)

# Log the figure to Comet
exp.log_figure(figure=fig)
exp.end()

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13

import comet_ml
import plotly.express as px

# Initialize the Comet Experiment
comet_ml.login()
exp = comet_ml.start()

# Create an example figure
df = px.data.gapminder().query("country=='Canada'")
fig = px.line(df, x="year", y="lifeExp", title='Life expectancy in Canada')

# Log the figure to Comet
exp.log_figure(figure=fig, figure_name="plotly_figure.html")

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13

import comet_ml
import seaborn as sns

# Initialize the Comet Experiment
comet_ml.login()
exp = comet_ml.start()

# Create an example line plot figure
fmri = sns.load_dataset("fmri")
line_plot = sns.lineplot(data=fmri, x="timepoint", y="signal", hue="event")

# Log the figure to Comet
exp.log_figure(figure=line_plot)

Additionally, you could add the figure_name argument to provide a descriptive name for the figure, use the overwrite argument to overwrite any existing asset with the same name, or pass the step argument to associate the figure to a training step.