Created
November 4, 2022 16:53
-
-
Save Polaris000/49d01caff4373542ba0eb6fa4facab78 to your computer and use it in GitHub Desktop.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| @jit | |
| def auc_recall_at_k_np_no_df_numba(y_true, y_conf): | |
| """ | |
| Experiment #4: | |
| -------------- | |
| Compute AUC under the Recall@k curve using numpy's | |
| functions. We do away with the conf_df dataframe | |
| as well. | |
| Numba's jit decorator is also added for further | |
| optimization. | |
| y_true: A numpy array of expected predictions | |
| y_conf: A numpy array of the model's confidence | |
| scores for each datapoint | |
| Returns: AUC-Recall@k (float) | |
| """ | |
| # if there are no positive targets (good leads), | |
| # auc becomes invalid | |
| if (y_true == 1).sum() == 0: | |
| return np.nan | |
| ranking = y_true[np.argsort(y_conf)[::-1]] | |
| # calculating recall@k based on sorted ranking | |
| recall_at_k = (ranking == 1).cumsum() / (ranking == 1).sum() | |
| # calculating ideal recall@k | |
| ideal_recall_at_k = np.minimum( | |
| np.ones(len(ranking)), | |
| np.array(list(range(1, len(ranking) + 1)))/ (ranking == 1).sum() | |
| ) | |
| return np.trapz(recall_at_k) / np.trapz(ideal_recall_at_k) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment