rhoa.plots

Visualization module for stock prediction analysis.

This module provides a pandas DataFrame accessor for creating publication-quality visualizations of stock prediction models. It includes methods for plotting buy/sell signals, confusion matrices, and performance metrics overlaid on price charts.

The visualizations are designed to help interpret model predictions, identify false positives/negatives, and understand prediction quality at a glance.

Examples

Basic usage with the DataFrame accessor:

>>> import pandas as pd
>>> import rhoa
>>> df = rhoa.read_csv('stock_data.csv')
>>> df['Date'] = pd.to_datetime(df['Date'])
>>> # Visualize predictions
>>> fig = df.rhoa.plots.signal(y_pred=predictions, y_true=targets)

Notes

All plotting methods return matplotlib Figure objects that can be further customized using standard matplotlib/seaborn APIs.

See also

rhoa.targets.future_return: Generate target labels based on future returns
rhoa.targets.drawdown: Generate target labels based on drawdown thresholds
matplotlib.pyplot.savefig: Save the figure to file
sklearn.metrics.confusion_matrix: Compute confusion matrix

Notes

Visualization Components:

When y_true is None (predictions only): - Single panel showing price chart with predicted buy signals (bright green) - No confusion matrix or error markers

When y_true is provided (validation mode): - Top panel: Confusion matrix with counts, percentages, and metrics - Bottom panel: Price chart with all signal types:

Light green background: True buy opportunities (ground truth)

Bright green dots: Model predicted buy signals

Red X markers: False positives (wrong predictions)

Orange circles: False negatives (missed opportunities)

Interpreting the Visualization:

Dense bright green clusters: Model is actively predicting buy signals
Green dots on light green background: True positives (correct predictions)
Red X markers: False alarms - model predicted buy but shouldn’t have
Orange circles: Missed trades - model failed to predict real opportunities
Gaps in signals: Periods where model predicts no buy opportunities

Best Practices:

Always provide y_true during model development for full diagnostics
Look for temporal patterns in false positives (e.g., during volatility)
Check if false negatives occur at specific price levels or market conditions
Compare precision/recall from confusion matrix with your trading strategy
Use threshold parameter to document the decision boundary
Save high-quality versions (dpi=300+) for documentation

Performance Metrics:

The confusion matrix panel displays: - Precision: Of all buy signals, what percentage were correct? - Recall: Of all true opportunities, what percentage did we catch? - Counts: Absolute numbers of TP, FP, TN, FN

Examples

Visualize predictions with full validation metrics:

>>> import pandas as pd
>>> import numpy as np
>>> import rhoa
>>>
>>> # Load stock data
>>> df = pd.read_csv('AAPL_stock_data.csv')
>>> df['Date'] = pd.to_datetime(df['Date'])
>>>
>>> # Generate targets using rhoa (7% return threshold)
>>> targets = df.targets.future_return(
...     threshold=0.07,
...     holding_period=10,
...     return_type='pct'
... )
>>>
>>> # Assume you have model predictions
>>> predictions = model.predict(features)
>>>
>>> # Create comprehensive visualization
>>> fig = df.rhoa.plots.signal(
...     y_pred=predictions,
...     y_true=targets,
...     threshold=0.67,
...     title='AAPL Random Forest Model',
...     save_path='outputs/aapl_predictions.png',
...     dpi=300
... )

Visualize predictions only (no ground truth available):

>>> # When you don't have labels (e.g., predicting future)
>>> fig = df.rhoa.plots.signal(
...     y_pred=predictions,
...     date_col='Date',
...     price_col='Close',
...     title='AAPL Future Predictions',
...     cmap='Greens'
... )

Customize for different price columns:

>>> # Use opening prices instead of closing
>>> fig = df.rhoa.plots.signal(
...     y_pred=predictions,
...     y_true=targets,
...     price_col='Open',
...     title='Entry Signals (Open Prices)'
... )

Save without displaying (batch processing):

>>> # Useful for generating reports for multiple stocks
>>> for ticker in ['AAPL', 'GOOGL', 'MSFT']:
...     df = load_stock_data(ticker)
...     predictions = model.predict(df)
...     fig = df.rhoa.plots.signal(
...         y_pred=predictions,
...         save_path=f'reports/{ticker}_signals.png',
...         show=False  # Don't display, just save
...     )
...     plt.close(fig)  # Free memory

Compare different thresholds visually:

>>> # Generate predictions at different thresholds
>>> proba = model.predict_proba(features)[:, 1]
>>>
>>> for thresh in [0.5, 0.67, 0.8]:
...     preds = (proba >= thresh).astype(int)
...     fig = df.rhoa.plots.signal(
...         y_pred=preds,
...         y_true=targets,
...         threshold=thresh,
...         title=f'Model Threshold {thresh}',
...         save_path=f'outputs/threshold_{thresh}.png',
...         show=False
...     )