Optimize F1 score in O(n^2)

100%

Optimize F1 score in O(n^2)

✓ Published0🌍 Public

Last edited Aug 3, 2017

Created on Aug 3, 2017

This visualization demonstrates the F1-score expectation maximization algorithm for optimizing predictions from posterior probabilities. It plots the expected F1 score against the number of positive labels (k), showing two curves for predictions with and without a "None" label, and marks the optimal k with a black dot. The code uses NumPy for dynamic programming calculations, pandas for data handling, and Matplotlib’s pyplot API with the ggplot style to render the line chart, annotations, and axis labels.

AI-generated description

Similar vizzes

Loading thumbnail…

Gist cfc8c25355ee6cf7e972cc74974aa369

The visualization plots the expected F1-score against the threshold k for a binary classification problem, comparing predictions that include a "None" label versus those without it. It uses a dynamic programming algorithm to compute expectations, then marks the optimal k with a black dot and annotation. The code generates the plot with pandas DataFrame plotting and matplotlib's pyplot, using the ggplot style, and also includes a separate benchmark chart showing quadratic runtime scaling for the expectation maximization process.

Aalvinthai

80% match

Loading thumbnail…

Gist 15ec54668ea15cb0eada554bae7d6f13

This example demonstrates the concept of precision at k by generating a synthetic dataset of 1,000 “potential bad actors,” each labeled as fraudulent or not, and then computing precision for the top 40 ranked predictions based on a model’s probability scores. The code uses `faker` to create fake names and emails, `numpy.random` for random labeling, and `scipy.stats.truncnorm` to simulate skewed probabilities. It defines a custom `precision_at_k` function that sorts scores in descending order and calculates the fraction of true positives among the first *k* entries, while also comparing it to scikit-learn’s `average_precision_score` for context.

BBrideau

55% match

Loading thumbnail…

prediction-scoring

This animated scatterplot demonstrates a prediction-scoring model, where each point represents a data instance evaluated by a scoring function. The visualization uses D3 v4 with SVG rendering to show points moving between predicted and actual values as scoring iterations progress. The animation highlights how predictions converge toward observed outcomes, with points colored and positioned by score error, updating in real time to reveal patterns in model performance across the distribution.

11wheel

53% match

Loading thumbnail…

prediction.html - Visualizing Uncertainty in Models using Pandas, Scikit Learn, Jinja, Flask, and D3.js

This example shows the prediction results of a machine learning model that classifies paper abstracts into IEEE Vis conference tracks (InfoVis, SciVis, VAST, or Vis). The interface displays the predicted category, the model’s accuracy, and bar charts for macro and weighted average precision, recall, and F1-score. The charts include an uncertainty value derived from the accuracy metric. Built with a Flask and Jinja backend, the page uses D3.js with a custom `uvis.barChart()` API that applies a gradient type, while the classification report is passed from Python’s scikit-learn into JavaScript for rendering.

SSaminu

52% match

Loading thumbnail…

Untitled0.ipynb

This notebook presents a markdown table comparing model performance under stretched versus untouched input conditions. The table reports a single row where a stretched dataset paired with a Gaussian probability distribution achieves an accuracy of 0.843. No code is executed; the visualization consists solely of the static table rendered in Markdown. The notebook uses no external plotting libraries or data-processing APIs, relying entirely on the notebook’s built-in Markdown rendering to display the comparison.

Aanemochore

49% match

Loading thumbnail…

Fork of proba

This example plots likelihood against a categorical severity score, connecting points with a red line to show the relationship between probability and impact. The y-axis uses a logarithmic scale, while the x-axis displays custom score labels. Built with D3 v5, it loads CSV data and renders an SVG using `scaleLinear`, `scaleLog`, `axisBottom`, `axisLeft`, and a line generator. Red circles mark each data point, and the chart includes a title and rotated axis label.

Aaleksstan96

49% match

Loading thumbnail…

Gist b6be7f1d29a9c01f70ec

This visualization shows the results of a permutation test for a Partial Least Squares (PLS) regression model predicting EDSS scores. It displays two side-by-side kernel density estimate plots of the distribution of $R^2$ values and root mean PRESS values across 1,000 random permutations, with vertical dashed lines marking the observed model performance (R² = 0.1282, rmPRESS = 1.6954). The empirical p-values computed from the permutation distributions are annotated directly on each plot. The chart uses matplotlib and seaborn, removes chartjunk, and displays the density plots with shaded colors, axis labels, and titles for each panel.

Aakeshavan

49% match

Loading thumbnail…

An implementation of Precision@k compatible with Scikit-learn.

This example demonstrates a custom implementation of the Precision@k metric that integrates directly with Scikit-learn's array utilities. It shows how to compute precision for the top *k* predictions, assuming a binary classification task. The function uses `column_or_1d` to flatten input arrays and `type_of_target` to validate the binary format. It sorts the true labels by descending prediction scores using `np.argsort`, then calculates the proportion of true positives among the top *k* entries. The code relies on NumPy for sorting and summing, and it returns a scalar float representing the precision value.

BBrideau

49% match