argueview-ArgueView是爲機器學習預測生成基於文本的演示文稿的工具，以及基於特徵重要性的解釋工具。該工具利用圖爾明的論證模型構建基於文本的解釋。-Sophia Hadash In...

argueview-ArgueView是爲機器學習預測生成基於文本的演示文稿的工具，以及基於特徵重要性的解釋工具。該工具利用圖爾明的論證模型構建基於文本的解釋。-Sophia Hadash

發佈:2020-12-22 17:31:32.174502

做者:Sophia Hadash

### 做者郵箱:s.hadash@tue.nl

### 首頁:https://github.com/SophiaHadash/ArgueView

### 文檔:None

### 下載連接

---

ArgueView is a tool for generating text-based presentations for machine-learning predictions and feature-importance based explanation tools. The tool makes use of Toulmin's model of argumentation for structuring the text-based explanations.

Example output using the visualizer:

The procedure for creating ArgueView explanations is as follows: 1. A traditional machine-learning context is created (i.e. data, model) 2. An explainer is employed to produce feature importance. This can be a white-box or black-box explainer. An example of a black-box explainer is LIME. 3. The machine-learning context and the feature importance are given to ArgueView such that it can produce a textual explanation.

Installation

ArgueView is available as a python package on PyPi. You can install it using pip:

pip install argueview

Or, using pipenv:

pipenv install argueview

Usage

Usage is documented in our examples. Examples are available in python and jupyter notebooks. The following examples are available:

• minimal, hypothetical data and explainer: python
• creditg data with LIME explainer: python, notebook
• creditg data with shap explainer: python, notebook

Running the examples

There are two examples available to help you learn how to use ArgueView. The 'plain' examples uses hypothetical data to show a minimalistic use-case. The CreditG example uses real data and a real ML model to illustrate a real-world use case.

If you would like to run the CreditG example the script needs to obtain the data. For this we use OpenML. However, usage requires a valid API key and you will need to obtain one to run the example.

After you have obtained your key, create a .env file with your OpenML API key.

echo "OML_APIKEY={my-key}" > .env

Note: You can skip this step if you want to run the hypothetical example.

Install all dependencies:

pipenv install --dev

Run an example:

/path/to/python3 ./examples/{example}/example.py

Additionally, there is are Jupyther Notebooks available to directly see how ArgueView works.

Building

Follow these steps to build ArgueView from source.

• make sure you install the dependencies. ArgueView requires the following dependencies: python>=3.5, pip3, pipenv, git.
• build using make make

Using Docker

Alternatively you can build ArgueView using docker.

• build context dockerfile docker build -t argueview/context:local .
• run build.sh in a container CID=$(docker run -v /var/run/docker.sock:/var/run/docker.sock argueview/context build.sh)

• copy results from the container

  docker cp ${CID}:/argueview/argueview.egg-info ./argueview.egg-info
  docker cp ${CID}:/argueview/build ./build
  docker cp ${CID}:/argueview/dist ./dist

  Copy from pypi.org

查詢時間：15.165ms

渲染時間：15.299ms

本文同步分享在 博客「zhenruyan」（other）。
若有侵權，請聯繫 support@oschina.cn 刪除。
本文參與「OSC源創計劃」，歡迎正在閱讀的你也加入，一塊兒分享。