The samay.visualization module provides visualization classes for different time series tasks.
ForecastVisualization
Visualize forecasting results with historical context, ground truth, and predictions.
from samay.visualization import ForecastVisualization
vis = ForecastVisualization(trues, preds, histories)
vis.plot(channel_idx=0, time_idx=0, crop_history=False)
Constructor
Historical context values
plot Method
Index of the channel (variate) to plot
Index of the time window to plot
If True, crops the history to 3x the forecast length. If False, shows the full history
Example
import numpy as np
from samay.visualization import ForecastVisualization
# Sample data
trues = [np.random.randn(10, 24) for _ in range(100)] # 100 windows, 10 channels, 24 timesteps
preds = [np.random.randn(10, 24) for _ in range(100)]
histories = [np.random.randn(10, 96) for _ in range(100)] # 96 historical timesteps
# Create visualization
vis = ForecastVisualization(trues, preds, histories)
# Plot specific channel and time window
vis.plot(channel_idx=0, time_idx=5, crop_history=True)
ImputationVisualization
Visualize imputation results showing ground truth, predictions, and missing value masks.
from samay.visualization import ImputationVisualization
vis = ImputationVisualization(trues, preds, masks)
vis.plot()
Constructor
Predicted (imputed) values
Binary mask indicating missing values (1 = missing, 0 = observed)
plot Method
Plots a randomly selected time window and channel, showing both the time series and the missing value mask.
Example
import numpy as np
from samay.visualization import ImputationVisualization
# Sample data (100 samples, 5 channels, 48 timesteps)
trues = np.random.randn(100, 5, 48)
preds = np.random.randn(100, 5, 48)
masks = np.random.randint(0, 2, size=(100, 5, 48)) # Binary mask
# Create and display visualization
vis = ImputationVisualization(trues, preds, masks)
vis.plot()
AnomalyDetectionVisualization
Visualize anomaly detection results with observed values, predictions, and anomaly scores.
from samay.visualization import AnomalyDetectionVisualization
vis = AnomalyDetectionVisualization(trues, preds, labels)
vis.plot()
Constructor
Binary labels indicating anomalies (1 = anomaly, 0 = normal)
plot Method
Displays a plot showing:
- Observed values (blue)
- Predicted values (red)
- Anomaly scores (black, calculated as squared error)
The plot focuses on a region around known anomalies (hardcoded indices: 74158-74984 with 512-step buffer).
Example
import numpy as np
from samay.visualization import AnomalyDetectionVisualization
# Sample data
trues = np.random.randn(100000)
preds = trues + np.random.randn(100000) * 0.1
labels = np.zeros(100000)
labels[74158:74984] = 1 # Mark anomaly region
# Create and display visualization
vis = AnomalyDetectionVisualization(trues, preds, labels)
vis.plot()
ClassificationVisualization
Visualize time series classification results using PCA dimensionality reduction of embeddings.
from samay.visualization import ClassificationVisualization
vis = ClassificationVisualization(embeddings, labels)
vis.plot()
Constructor
Time series embeddings (feature vectors)
Class labels for each embedding
plot Method
Creates a 2D scatter plot of embeddings after PCA dimensionality reduction, colored by class labels.
Example
import numpy as np
from samay.visualization import ClassificationVisualization
# Sample embeddings (500 samples, 128 features)
embeddings = np.random.randn(500, 128)
labels = np.random.randint(0, 5, size=500) # 5 classes
# Create and display visualization
vis = ClassificationVisualization(embeddings, labels)
vis.plot()
Legacy Visualization Function
The module also provides a legacy visualize function from samay.utils for backward compatibility.
from samay.utils import visualize
visualize(
task_name="forecasting",
trues=trues,
preds=preds,
history=history,
context_len=512,
dataset="ETTh1",
freq="1H"
)
visualize function.
Usage Notes
Task-Specific Visualizations
Each visualization class is designed for a specific task:
- ForecastVisualization: Shows historical context, ground truth, and forecast
- ImputationVisualization: Shows missing value masks and imputed values
- AnomalyDetectionVisualization: Shows anomaly scores alongside observations
- ClassificationVisualization: Shows embedding space in 2D using PCA
Customization
All visualization classes use matplotlib under the hood. You can customize plots by:
- Modifying the plot after calling
.plot()
- Using
plt.figure() before calling .plot() to set figure size
- Adding additional plot elements after
.plot() returns
import matplotlib.pyplot as plt
from samay.visualization import ForecastVisualization
plt.figure(figsize=(16, 6))
vis = ForecastVisualization(trues, preds, histories)
vis.plot(channel_idx=0, time_idx=0)
plt.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig("forecast.png", dpi=300)