👩‍💻 Welcome to PyExplainer Quickstart Guide 👨‍💻 ¶

🛠 Installation¶

- Please ignore this part if you cloned the whole package from GitHub¶

🤖 Run the cell below to install pyexplainer 0.1.5¶

In [ ]:

!pip install pyexplainer

🤖 If the code above did not work, try the cell below, otherwise, you are good to go!¶

In [ ]:

!pip3 install pyexplainer

Let's get started !¶

👩🏻‍🔧 1. Prepare data and model¶

📝Note. We use the default data and model here for an example¶

1.1 Import Libraries Needed¶

In [2]:

from pyexplainer import pyexplainer_pyexplainer
from pyexplainer.pyexplainer_pyexplainer import PyExplainer 

1.2 Use default datasets and model (Random Forest)¶

In [3]:

default_data_and_model = pyexplainer_pyexplainer.get_default_data_and_model()
py_explainer = PyExplainer(X_train = default_data_and_model['X_train'],
                           y_train = default_data_and_model['y_train'],
                           indep = default_data_and_model['indep'],
                           dep = default_data_and_model['dep'],
                           blackbox_model = default_data_and_model['blackbox_model'])

🔧2. Create a Rule Object Manually¶

📝Note. Rule Object is the core backend concept of PyExplainer !¶

2.1 Prepare X_explain and y_explain data¶

📝Note. We use the default data here for an example¶

In [4]:

X_explain = default_data_and_model['X_explain']

In [5]:

y_explain = default_data_and_model['y_explain']

2.2 Create the rule object¶

In [6]:

created_rule_obj = py_explainer.explain(X_explain=X_explain,
                                        y_explain=y_explain,
                                        search_function='crossoverinterpolation')

👩🏽‍🎨 3. Pass Rule Object to .visualise(rule_obj) to Generate the Bullet Chart and Interactive Slider¶

📝Note. simply move the gray slider to modify the value so you can get a new risk score.¶

🔧 Visualise the Rule Object we created manually using .explain(...) method¶

In [7]:

py_explainer.visualise(created_rule_obj)

HBox(children=(Label(value='Risk Score: '), FloatProgress(value=0.0, bar_style='info', layout=Layout(width='40…

FloatSlider(value=0.0, continuous_update=False, description='#1 Decrease the values of CountDeclMethodDefault …

FloatSlider(value=1.54, continuous_update=False, description='#2 Increase the values of RatioCommentToCode to …

FloatSlider(value=1.0, continuous_update=False, description='#3 Decrease the values of AvgCyclomaticModified t…

Output(layout=Layout(border='3px solid black'))

🤡 Important - Bug Report Channel 🤡¶

Please report here ¶

📧 or email your report to michaelfu1998@gmail.com¶

¶

🙏Thanks for playing around with PyExplainer, I really appreciate your time! 🙏 ¶

🔥 More Features will be Released Soon 🔥 ¶

📜 Appendex 📜¶

A. 🕵🏻 What's in the Rule Object (rule_obj) ? Let's unbox it ! 📦¶

Basic Data Check¶

In [ ]:

print("Type of Rule Object: ", type(load_pyExp_rule_obj))
print()
print("All of the keys in Rule Object")
i = 1
for k in load_pyExp_rule_obj.keys():
    print("Key ", i, " - ",k)
    i += 1

🔑 Key 1 - synthetic_data¶

As can be seen below, the synthetic data are data coming from feature columns¶

This synthetic data was generated internally by the PyExplainer when the .explain(...) method is triggered¶

Currently we have 2 approaches to generate synthetic_data¶

Approach (1) Crossover and Interpolation¶

Approach (2) Random Pertubation¶

After the process of C&I. or RP., synthetic_data will be generated as a DataFrame below¶

In [ ]:

print("Type of pyExp_rule_obj['synthetic_data'] - ", type(load_pyExp_rule_obj['synthetic_data']), "\n")
print("Example", "\n\n", load_pyExp_rule_obj['synthetic_data'].head(2))

🔑 Key 2 - synthetic_predictions¶

As can be seen below, the synthetic prediction are data coming from the prediction column¶

This synthetic prediction was generated internally by the PyExplainer when the .explain(...) method is triggered¶

This synthetic prediction is created based on the black box model we passed to the PyExplainer when initialising (section 1.5 & 2.3)¶

This synthetic prediction is generated based on the synthetic data above therefore it's called synthetic_predictions¶

>>> e.g. synthetic_predictions = blackbox_model.predict(synthetic_data) Note. we only need feature cols in synthetic_data¶

In [ ]:

print("Type of pyExp_rule_obj['synthetic_predictions'] - ", type(load_pyExp_rule_obj['synthetic_predictions']), "\n")
print("Example", "\n\n", load_pyExp_rule_obj['synthetic_predictions'])

🔑 Key 3 - X_explain¶

This X_explain is exactly the same as the one we passed to .explain(...) method (section 3.3 & section 3.4)¶

In [ ]:

print("Type of pyExp_rule_obj['X_explain'] - ", type(load_pyExp_rule_obj['X_explain']), "\n")
print("Example", "\n\n", load_pyExp_rule_obj['X_explain'])

🔑 Key 4 - y_explain¶

This y_explain is exactly the same as the one we passed to .explain(...) method (section 3.3 & section 3.4)¶

In [ ]:

print("Type of pyExp_rule_obj['y_explain'] - ", type(load_pyExp_rule_obj['y_explain']), "\n")
print("Example", "\n\n", load_pyExp_rule_obj['y_explain'])

🔑 Key 5 - indep¶

Names of the Selected Feature Cols¶

In [ ]:

print("Type of pyExp_rule_obj['indep'] - ", type(load_pyExp_rule_obj['indep']), "\n")
print("Example", "\n\n", load_pyExp_rule_obj['indep'])

🔑 Key 6 - dep¶

Names of the Label Col (Prediction Col)¶

In [ ]:

print("Type of pyExp_rule_obj['dep'] - ", type(load_pyExp_rule_obj['dep']), "\n")
print("Example", "\n\n", load_pyExp_rule_obj['dep'])

🔑 Key 7 - top_k_positive_rules¶

This shows the top k positive rules generated by the RuleFit model inside the .explain(...) function¶

The value of 'top_k' can be tuned in when we create a Rule Object manually (section 3.4), the default value is 3¶

In [ ]:

print("Type of pyExp_rule_obj['top_k_positive_rules'] - ", type(load_pyExp_rule_obj['top_k_positive_rules']), "\n")
print("Example", "\n\n", load_pyExp_rule_obj['top_k_positive_rules'])

🔑 Key 8 - top_k_negative_rules¶

This shows the top k negative rules generated by the RuleFit model inside the .explain(...) function¶

The value of 'top_k' can be tuned in when we create a Rule Object manually (section 3.4), the default value is 3¶

However, in the current version, the top_k value is always the same for both negative and positive rules which can be improved in the future version¶

In [ ]:

print("Type of pyExp_rule_obj['top_k_negative_rules'] - ", type(load_pyExp_rule_obj['top_k_negative_rules']), "\n")
print("Example", "\n\n", load_pyExp_rule_obj['top_k_negative_rules'])

👩‍💻 Welcome to PyExplainer Quickstart Guide 👨‍💻 ¶

pyexplainer - GitHub Repository¶

pyexplainer - PyPI¶

pyexplainer - Official Documentation¶

🛠 Installation¶

- Please ignore this part if you cloned the whole package from GitHub¶

🤖 Run the cell below to install pyexplainer 0.1.5¶

🤖 If the code above did not work, try the cell below, otherwise, you are good to go!¶

Let's get started !¶

👩🏻‍🔧 1. Prepare data and model¶

📝Note. We use the default data and model here for an example¶

1.1 Import Libraries Needed¶

1.2 Use default datasets and model (Random Forest)¶

🔧2. Create a Rule Object Manually¶

📝Note. Rule Object is the core backend concept of PyExplainer !¶

2.1 Prepare X_explain and y_explain data¶

📝Note. We use the default data here for an example¶

2.2 Create the rule object¶

👩🏽‍🎨 3. Pass Rule Object to .visualise(rule_obj) to Generate the Bullet Chart and Interactive Slider¶

📝Note. simply move the gray slider to modify the value so you can get a new risk score.¶

🔧 Visualise the Rule Object we created manually using .explain(...) method¶

🤡 Important - Bug Report Channel 🤡¶

Please report here¶

📧 or email your report to michaelfu1998@gmail.com¶

¶

🙏Thanks for playing around with PyExplainer, I really appreciate your time! 🙏 ¶

🔥 More Features will be Released Soon 🔥 ¶

📜 Appendex 📜¶

A. 🕵🏻 What's in the Rule Object (rule_obj) ? Let's unbox it ! 📦¶

Basic Data Check¶

🔑 Key 1 - synthetic_data¶

As can be seen below, the synthetic data are data coming from feature columns¶

This synthetic data was generated internally by the PyExplainer when the .explain(...) method is triggered¶

Currently we have 2 approaches to generate synthetic_data¶

Approach (1) Crossover and Interpolation¶

Approach (2) Random Pertubation¶

After the process of C&I. or RP., synthetic_data will be generated as a DataFrame below¶

🔑 Key 2 - synthetic_predictions¶

As can be seen below, the synthetic prediction are data coming from the prediction column¶

This synthetic prediction was generated internally by the PyExplainer when the .explain(...) method is triggered¶

This synthetic prediction is created based on the black box model we passed to the PyExplainer when initialising (section 1.5 & 2.3)¶

This synthetic prediction is generated based on the synthetic data above therefore it's called synthetic_predictions¶

>>> e.g. synthetic_predictions = blackbox_model.predict(synthetic_data) Note. we only need feature cols in synthetic_data¶

🔑 Key 3 - X_explain¶

This X_explain is exactly the same as the one we passed to .explain(...) method (section 3.3 & section 3.4)¶

🔑 Key 4 - y_explain¶

This y_explain is exactly the same as the one we passed to .explain(...) method (section 3.3 & section 3.4)¶

🔑 Key 5 - indep¶

Names of the Selected Feature Cols¶

🔑 Key 6 - dep¶

Names of the Label Col (Prediction Col)¶

🔑 Key 7 - top_k_positive_rules¶

This shows the top k positive rules generated by the RuleFit model inside the .explain(...) function¶

The value of 'top_k' can be tuned in when we create a Rule Object manually (section 3.4), the default value is 3¶

🔑 Key 8 - top_k_negative_rules¶

This shows the top k negative rules generated by the RuleFit model inside the .explain(...) function¶

The value of 'top_k' can be tuned in when we create a Rule Object manually (section 3.4), the default value is 3¶

However, in the current version, the top_k value is always the same for both negative and positive rules which can be improved in the future version¶

Please report here ¶