## Heatmap: Accuracy ### Overview The image is a heatmap displaying the accuracy between two digits, 'digit_a' and 'digit_b', ranging from 0.0 to 1.0. The heatmap is triangular, with values only present in the upper triangle. The color intensity represents the accuracy, with darker blues indicating higher accuracy and darker reds indicating lower accuracy. ### Components/Axes * **Title:** Accuracy * **X-axis:** digit\_a (values: 1, 2, 3, 4, 5) * **Y-axis:** digit\_b (values: 1, 2, 3, 4, 5) * **Colorbar:** Ranges from 0.0 (dark red) to 1.0 (dark blue), with intermediate values indicated by a gradient. The colorbar has tick marks at 0.0, 0.2, 0.4, 0.6, 0.8, and 1.0. ### Detailed Analysis or ### Content Details The heatmap displays the following accuracy values for each digit pair: * **digit\_b = 1:** * digit\_a = 1: 1.00 (dark blue) * **digit\_b = 2:** * digit\_a = 1: 1.00 (dark blue) * digit\_a = 2: 0.98 (dark blue) * **digit\_b = 3:** * digit\_a = 1: 0.99 (dark blue) * digit\_a = 2: 0.18 (red) * digit\_a = 3: 0.11 (red) * **digit\_b = 4:** * digit\_a = 1: 0.92 (blue) * digit\_a = 2: 0.11 (red) * digit\_a = 3: 0.09 (red) * digit\_a = 4: 0.03 (dark red) * **digit\_b = 5:** * digit\_a = 1: 0.09 (red) * digit\_a = 2: 0.11 (red) * digit\_a = 3: 0.10 (red) * digit\_a = 4: 0.09 (red) * digit\_a = 5: 0.05 (red) ### Key Observations * The accuracy is highest when digit\_a and digit\_b are the same or close to each other, especially for digits 1 and 2. * The accuracy drops significantly when comparing digits 1 and 2 to digits 3, 4, and 5. * The lowest accuracy values are observed when comparing digit 5 to other digits. ### Interpretation The heatmap suggests that the model performs well in distinguishing between digits 1 and 2, but struggles with digits 3, 4, and 5. There is a clear distinction in accuracy between the lower digits (1 and 2) and the higher digits (3, 4, and 5). This could indicate that the features used to classify the digits are more effective for the lower digits or that there is more overlap in the features for the higher digits. The poor performance when comparing digit 5 to other digits suggests that digit 5 may be particularly difficult to classify.

## Heatmap: Accuracy ### Overview This image presents a heatmap visualizing the accuracy of a digit recognition system, specifically showing how well the system confuses one digit for another. The heatmap displays accuracy values ranging from 0.0 to 1.0, with darker blues representing higher accuracy and darker reds representing lower accuracy. The axes represent the 'true' digit (digit_a) and the 'predicted' digit (digit_b). ### Components/Axes * **Title:** "Accuracy" - positioned at the top-center of the image. * **X-axis Label:** "digit\_a" - representing the actual digit. Markers are 1, 2, 3, 4, 5. * **Y-axis Label:** "digit\_b" - representing the predicted digit. Markers are 1, 2, 3, 4, 5. * **Colorbar:** Located on the right side of the heatmap. It ranges from 0.0 (dark red) to 1.0 (dark blue), representing the accuracy scale. * **Data Cells:** Each cell in the grid represents the accuracy of predicting digit 'b' when the actual digit is 'a'. The values are displayed within each cell. ### Detailed Analysis The heatmap is a 5x5 grid. I will analyze row by row, referencing the colorbar to confirm accuracy. * **Row 1 (digit\_a = 1):** * digit\_b = 1: 1.00 (dark blue) * digit\_b = 2: 1.00 (dark blue) * digit\_b = 3: 0.99 (light blue) * digit\_b = 4: 0.92 (medium blue) * digit\_b = 5: 0.09 (dark red) * **Row 2 (digit\_a = 2):** * digit\_b = 1: 1.00 (dark blue) * digit\_b = 2: 0.98 (light blue) * digit\_b = 3: 0.18 (red) * digit\_b = 4: 0.11 (red) * digit\_b = 5: 0.09 (dark red) * **Row 3 (digit\_a = 3):** * digit\_b = 1: 0.99 (light blue) * digit\_b = 2: 0.18 (red) * digit\_b = 3: 0.11 (red) * digit\_b = 4: 0.09 (dark red) * digit\_b = 5: 0.05 (dark red) * **Row 4 (digit\_a = 4):** * digit\_b = 1: 0.92 (medium blue) * digit\_b = 2: 0.11 (red) * digit\_b = 3: 0.09 (dark red) * digit\_b = 4: 0.03 (dark red) * digit\_b = 5: 0.09 (dark red) * **Row 5 (digit\_a = 5):** * digit\_b = 1: 0.09 (dark red) * digit\_b = 2: 0.11 (red) * digit\_b = 3: 0.10 (red) * digit\_b = 4: 0.09 (dark red) * digit\_b = 5: 0.05 (dark red) ### Key Observations * The system performs very well at recognizing digits 1 and 2 when they are the actual digits. Accuracy is near 1.0 when predicting 1 or 2 given 1 or 2. * The system frequently confuses digit 5 for other digits, and vice versa. Accuracy values are consistently low (around 0.05-0.11) when the actual digit is 5. * Digit 4 is also often misclassified, particularly as other digits. * The diagonal elements (where predicted = actual) are the highest values, as expected. * There is a noticeable asymmetry in the confusion matrix. For example, the system is better at predicting '1' when the actual digit is '1' than it is at predicting '1' when the actual digit is '5'. ### Interpretation This heatmap demonstrates the performance of a digit recognition model. The high accuracy along the diagonal indicates that the model correctly identifies digits most of the time when the input is clear. However, the low accuracy values in several off-diagonal cells reveal specific areas of confusion. The model struggles to differentiate between digits 5 and other digits, suggesting that the features used to recognize '5' are similar to those of other digits, or that the training data for '5' was insufficient or of poor quality. The heatmap provides valuable insights for improving the model, such as focusing on feature engineering or data augmentation for the digits that are frequently misclassified. The pattern of confusion suggests that the model may be sensitive to certain visual characteristics of the digits, leading to systematic errors.

## Heatmap: Accuracy ### Overview The image is a heatmap visualizing the accuracy of digit comparisons between two variables, `digit_a` (x-axis) and `digit_b` (y-axis). Accuracy values range from 0.0 to 1.0, with blue representing higher accuracy and red representing lower accuracy. The diagonal (where `digit_a` = `digit_b`) shows the highest accuracies, while off-diagonal values decrease as digits diverge. ### Components/Axes - **Title**: "Accuracy" - **X-axis**: Labeled "digit_a" with categories 1–5. - **Y-axis**: Labeled "digit_b" with categories 1–5. - **Legend**: Positioned on the right, showing a gradient from dark blue (1.0) to red (0.0). ### Detailed Analysis | digit_b \ digit_a | 1 | 2 | 3 | 4 | 5 | |-------------------|-------|-------|-------|-------|-------| | **1** | 1.00 | 1.00 | 0.99 | 0.92 | 0.09 | | **2** | 1.00 | 0.98 | 0.18 | 0.11 | 0.03 | | **3** | 0.99 | 0.18 | 0.11 | 0.09 | 0.05 | | **4** | 0.92 | 0.11 | 0.09 | 0.03 | 0.05 | | **5** | 0.09 | 0.11 | 0.10 | 0.09 | 0.05 | - **Diagonal Values**: - `digit_a` = `digit_b` = 1: 1.00 (dark blue) - `digit_a` = `digit_b` = 2: 0.98 (dark blue) - `digit_a` = `digit_b` = 3: 0.99 (dark blue) - `digit_a` = `digit_b` = 4: 0.92 (light blue) - `digit_a` = `digit_b` = 5: 0.05 (dark red) — **anomaly** (extremely low accuracy for matching digits). - **Off-Diagonal Trends**: - Accuracy decreases as `digit_a` and `digit_b` differ, especially for non-adjacent digits (e.g., 1 vs 5: 0.09). - Adjacent digits (e.g., 1 vs 2: 1.00, 2 vs 3: 0.18) show mixed performance, with some high and some low values. ### Key Observations 1. **Diagonal Dominance**: The highest accuracies occur when `digit_a` and `digit_b` are identical, except for `digit_b` = 5, which has a critical outlier (0.05). 2. **Confusion Between Digits**: - Digits 1 and 2 show near-perfect accuracy when compared to themselves but drop sharply when mismatched (e.g., 1 vs 5: 0.09). - Digit 5 exhibits poor performance overall, with low accuracy even when compared to itself. 3. **Color Gradient Consistency**: Blue shades align with high values (1.0–0.92), while red shades correspond to low values (0.11–0.05). ### Interpretation The heatmap suggests a classification model optimized for exact digit matches but struggles with distinguishing dissimilar digits, particularly digit 5. The anomaly for `digit_b` = 5 (0.05 accuracy when matched) indicates a potential flaw in the model’s handling of this digit, possibly due to data imbalance, feature extraction issues, or misclassification during training. The sharp drop in accuracy for non-adjacent digits (e.g., 1 vs 5) highlights challenges in generalizing across visually distinct classes. This pattern is typical in digit recognition tasks, where confusion matrices often reveal class-specific weaknesses.