InterpretML

https://interpret.ml/docs
Admin
InterpretML is an open-source toolkit for machine learning interpretability that enables training...
Tokens:150,692
Snippets:1,231
Trust Score:9.3
Update:1 day ago
Show doc for...
Tokens:
Raw
### Install interpret with every dependency

Source: https://interpret.ml/docs/installation-guide.html

Builds the project, installs JavaScript dependencies, and installs the Python package with all optional dependencies.

```bash
cd interpret

./build.sh  # OR use build.bat on Windows

cd shared/vis

npm run clean && npm install && npm run build-prod

cd ../../python/interpret-core

pip install -e .[debug,notebook,plotly,lime,sensitivity,shap,linear,skoperules,treeinterpreter,aplr,dash,testing]
```

--------------------------------

### Install interpret with every dependency (source)

Source: https://interpret.ml/docs/deployment-guide.html

Clones the interpret repository and runs the install script to install the full package from source.

```bash
git clone https://github.com/interpretml/interpret.git && cd interpret/scripts && make install
```

--------------------------------

### Install interpret with all dependencies (pip, conda, source)

Source: https://interpret.ml/docs/_sources/deployment-guide.ipynb

Use these commands to install the full `interpret` package, which includes all dependencies for visualization, glassbox models, and blackbox explanations. The pip command installs from PyPI, the conda command installs from conda-forge, and the source command clones the repository and runs the install script.

```shell
pip install interpret
```

```shell
conda install -c conda-forge interpret
```

```shell
git clone https://github.com/interpretml/interpret.git && cd interpret/scripts && make install
```

--------------------------------

### Install interpret with every dependency (pip)

Source: https://interpret.ml/docs/deployment-guide.html

Installs the full `interpret` package with all dependencies via pip. This is the default installation method.

```bash
pip install interpret
```

--------------------------------

### Install interpret-core with minimal dependencies (source)

Source: https://interpret.ml/docs/deployment-guide.html

Clones the interpret repository and runs the install-core script to install the minimal package from source.

```bash
git clone https://github.com/interpretml/interpret.git && cd interpret/scripts && make install-core
```

--------------------------------

### Install interpret-core with all extra tags (pip)

Source: https://interpret.ml/docs/deployment-guide.html

Installs `interpret-core` with all extra tags, which is equivalent to installing the full `interpret` package. This is only supported via pip.

```bash
pip install interpret-core[debug,notebook,plotly,lime,sensitivity,shap,linear,skoperules,treeinterpreter,aplr,dash,testing]
```

--------------------------------

### Install interpret and dependencies

Source: https://interpret.ml/docs/python/examples/explain-blackbox-regressors.html

Installs interpret, pandas, scikit-learn, and lime if interpret is not already installed. Uses pip with the --quiet flag.

```python
# install interpret if not already installed
try:
    import interpret
except ModuleNotFoundError:
    !pip install --quiet interpret pandas scikit-learn lime
```

--------------------------------

### VS Code launch configurations for Python and C++ Attach

Source: https://interpret.ml/docs/debugging-guide.html

Example launch.json configurations for attaching to a C++ process and launching the current Python file.

```json
"configurations": [
    {
        "name": "(Windows) Attach",
        "type": "cppvsdbg",
        "request": "attach",
        "processId": "${command:pickProcess}"
    },
    {
        "name": "Python: Current File",
        "type": "python",
        "request": "launch",
        "program": "${file}",
        "console": "integratedTerminal"
    }
]
```

--------------------------------

### Install interpret-core with minimal dependencies (conda)

Source: https://interpret.ml/docs/deployment-guide.html

Installs the minimal `interpret-core` package via conda from conda-forge.

```bash
conda install -c conda-forge interpret-core
```

--------------------------------

### Show a single explanation with interpret.show

Source: https://interpret.ml/docs/_sources/framework.ipynb

This example loads the Adult dataset, trains an ExplainableBoostingClassifier, and calls show to render a dropdown widget with an overview and per-feature graphs for a global explanation.

```python
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split

from interpret.glassbox import ExplainableBoostingClassifier
from interpret import show

df = pd.read_csv(
    "https://archive.ics.uci.edu/ml/machine-learning-databases/adult/adult.data",
```

--------------------------------

### Install interpret and dependencies

Source: https://interpret.ml/docs/python/examples/differential-privacy.html

Installs the interpret, pandas, and scikit-learn packages if interpret is not already installed.

```python
try:
    import interpret
except ModuleNotFoundError:
    !pip install --quiet interpret pandas scikit-learn
```

--------------------------------

### Install interpret and scikit-learn if not already installed

Source: https://interpret.ml/docs/python/examples/interpretable-regression.html

Checks if the interpret package is importable; if not, installs interpret and scikit-learn using pip. This is a prerequisite for running the notebook.

```python
# install interpret if not already installed
try:
    import interpret
except ModuleNotFoundError:
    !pip install --quiet interpret scikit-learn
```

--------------------------------

### Install interpret-core with minimal dependencies (pip)

Source: https://interpret.ml/docs/deployment-guide.html

Installs the minimal `interpret-core` package via pip, which includes only dependencies required for EBM fitting, predicting, editing, serialization, and generating explanations.

```bash
pip install interpret-core
```

--------------------------------

### Install interpret-core with minimal dependencies (pip, conda, source)

Source: https://interpret.ml/docs/_sources/deployment-guide.ipynb

Use these commands to install the `interpret-core` package, which includes only the dependencies required for EBM fitting, predicting, editing, serialization, and generating explanations. This is suitable for production deployments with fewer dependencies.

```shell
pip install interpret-core
```

```shell
conda install -c conda-forge interpret-core
```

```shell
git clone https://github.com/interpretml/interpret.git && cd interpret/scripts && make install-core
```

--------------------------------

### Import boilerplate for EBM multiclass example

Source: https://interpret.ml/docs/ebm-internals-multiclass.html

Imports required libraries: interpret.show, ExplainableBoostingClassifier, and numpy.

```python
# boilerplate
from interpret import show
from interpret.glassbox import ExplainableBoostingClassifier
import numpy as np
```

--------------------------------

### Install interpret with every dependency (conda)

Source: https://interpret.ml/docs/deployment-guide.html

Installs the full `interpret` package with all dependencies via conda from conda-forge.

```bash
conda install -c conda-forge interpret
```

--------------------------------

### Build a Blackbox Pipeline

Source: https://interpret.ml/docs/index.html

Example of creating a scikit-learn Pipeline with PCA and RandomForestClassifier for blackbox interpretability.

```APIDOC
## Build a Blackbox Pipeline

### Description
Construct a blackbox machine learning pipeline that includes preprocessing steps (PCA) and a classifier (RandomForestClassifier) for interpretability.

### Code Example
```python
from sklearn.ensemble import RandomForestClassifier
from sklearn.decomposition import PCA
from sklearn.pipeline import Pipeline

# We have to transform categorical variables to use sklearn models
X = pd.get_dummies(X, prefix_sep='.').astype(float)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20, random_state=seed)

#Blackbox system can include preprocessing, not just a classifier!
pca = PCA()
rf = RandomForestClassifier(random_state=seed)

blackbox_model = Pipeline([('pca', pca), ('rf', rf)])
blackbox_model.fit(X_train, y_train)
```

### Pipeline Parameters
- **steps** (list of tuples) - Required - List of (name, estimator) tuples to be chained in sequential order. All steps must define `fit`; non-last steps must define `transform`.
- **transform_input** (list of str, default=None) - Optional - Names of metadata parameters to be transformed by the pipeline before passing to the step consuming it. Requires metadata routing enabled.
- **memory** (str or object with joblib.Memory interface, default=None) - Optional - Used to cache fitted transformers. If string, path to caching directory.
- **verbose** (bool, default=False) - Optional - If True, prints time elapsed while fitting each step.

### PCA Parameters
- **n_components** (int, float or 'mle', default=None) - Optional - Number of components to keep. If not set, all components are kept (min(n_samples, n_features)). If 'mle' and svd_solver='full', Minka's MLE is used. If 0 < n_components < 1 and svd_solver='full', selects number of components to explain variance greater than the percentage specified.
```

--------------------------------

### Show multiple explanations for comparison with show([ebm_global, lr_global])

Source: https://interpret.ml/docs/framework.html

Pass a list of explanation objects to the show function to display a comparison dashboard. This example compares the EBM global explanation with the Logistic Regression global explanation.

```python
show([ebm_global, lr_global])
```

--------------------------------

### interpret.init_show_server(addr=None, base_url=None, use_relative_links=False)

Source: https://interpret.ml/docs/python/api/init_show_server.html

Initializes the show method's backing server. This function allows you to configure the server address, base URL path, and whether to use relative links in responses.

```APIDOC
## interpret.init_show_server(addr=None, base_url=None, use_relative_links=False)

### Description
Initializes the show method's backing server. This function allows you to configure the server address, base URL path, and whether to use relative links in responses.

### Parameters
- **addr** (tuple) - Optional - (ip, port) tuple as address to assign show method to.
- **base_url** (str) - Optional - Base url path as string. Used mostly when server is running behind a proxy.
- **use_relative_links** (bool) - Optional - Use relative links for what's returned to client. Otherwise have absolute links.
```

--------------------------------

### Install interpret if not already installed

Source: https://interpret.ml/docs/python/examples/prototype-selection.html

Checks if the interpret package is available and installs it along with numpy, scikit-learn, and matplotlib if not. Uses a try/except block and a pip install command.

```python
# install interpret if not already installed
try:
    import interpret
except ModuleNotFoundError:
    !pip install --quiet interpret numpy scikit-learn matplotlib
```

--------------------------------

### Install interpret if not already installed

Source: https://interpret.ml/docs/python/examples/custom-interactions.html

Checks if the interpret package is importable; if not, installs interpret, pandas, and scikit-learn using pip.

```python
# install interpret if not already installed
try:
    import interpret
except ModuleNotFoundError:
    !pip install --quiet interpret pandas scikit-learn
```

--------------------------------

### LimeTabular.__init__

Source: https://interpret.ml/docs/python/api/LimeTabular.html

Initializes the LimeTabular explainer with a model, data, and optional feature names and types.

```APIDOC
## LimeTabular.__init__

### Description
Initializes the LimeTabular class, exposing the LIME tabular explainer from the lime package in interpret API form.

### Method
Constructor

### Signature
_class_ interpret.blackbox.LimeTabular(_model_, _data_, _feature_names=None_, _feature_types=None_, _**_kwargs_)

### Parameters
- **model** - Required - Model or prediction function of model (predict_proba for classification or predict for regression)
- **data** - Required - Data used to initialize LIME with.
- **feature_names** - Optional - List of feature names.
- **feature_types** - Optional - List of feature types.
- **kwargs** - Optional - Kwargs that will be sent to lime
```

--------------------------------

### interpret.init_show_server

Source: https://interpret.ml/docs/_sources/python/api/init_show_server.ipynb

Initializes the show server for the interpret library. This function is part of the interpret package's public API.

```APIDOC
## interpret.init_show_server

### Description
Initializes the show server for the interpret library.

### Method
Function call

### Endpoint
interpret.init_show_server

### Parameters
No parameters documented.

### Request Example
```python
import interpret
interpret.init_show_server()
```

### Response
No return value documented.
```

--------------------------------

### Output of intercept_

Source: https://interpret.ml/docs/ebm-internals-regression.html

Example output showing the intercept value, which is 450.0 in this example.

```text
450.0
```

--------------------------------

### Generate full interpret dashboard with show()

Source: https://interpret.ml/docs/faq.html

Pass a list of explanations to show() to produce the full interpret dashboard. Passing a single explanation (not wrapped in a list) returns a small dropdown instead.

```python
from interpret import show

show(ebm_local)   # Returns small dropdown
show([ ebm_local ]) # Produces interpret dashboard
```

--------------------------------

### Install EBM without other explainers using interpret-core

Source: https://interpret.ml/docs/faq.html

Use this pip command to install only the EBM explainer from the interpret-core base package, avoiding other dependencies. The 'required' tag is generally recommended for all installs.

```bash
pip install interpret-core[required,ebm]
```

--------------------------------

### LimeTabular.explain_local

Source: https://interpret.ml/docs/python/api/LimeTabular.html

Generates local explanations for provided instances using LIME.

```APIDOC
## LimeTabular.explain_local

### Description
Generates local explanations for provided instances.

### Method
explain_local(_X_, _y=None_, _name=None_, _**_kwargs_)

### Parameters
- **X** - Required - Numpy array for X to explain.
- **y** - Optional - Numpy vector for y to explain.
- **name** - Optional - User-defined explanation name.
- **kwargs** - Optional - Kwargs that will be sent to lime

### Returns
An explanation object, visualizing feature-value pairs for each instance as horizontal bar charts.
```

--------------------------------

### Local explanation with ShapKernel

Source: https://interpret.ml/docs/python/examples/explain-blackbox-regressors.html

Creates a SHAP Kernel local explainer. Requires a background dataset (here, the median of X_train) and a blackbox model. Explains the first five test instances and displays the explanation for the first instance.

```python
from interpret.blackbox import ShapKernel

background_val = pd.DataFrame(np.median(X_train, axis=0).reshape(1, -1), columns=X.columns)
shap = ShapKernel(blackbox_model, background_val)
shap_local = shap.explain_local(X_test[:5], y_test[:5], name='SHAP')
show(shap_local, 0)
```

--------------------------------

### explain_local(X, y=None, name=None, init_score=None)

Source: https://interpret.ml/docs/python/api/ExplainableBoostingRegressor.html

Provides local explanations for provided samples.

```APIDOC
## explain_local(X, y=None, name=None, init_score=None)

### Description
Provide local explanations for provided samples.

### Parameters
- **X** - Required - NumPy array for X to explain.
- **y** - Optional - NumPy vector for y to explain.
- **name** - Optional - User-defined explanation name.
- **init_score** - Optional - Initial score.
```

--------------------------------

### get_num_cv_folds

Source: https://interpret.ml/docs/python/api/APLRRegressor.html

Gets the number of cross-validation folds used during training.

```APIDOC
## get_num_cv_folds

### Description
Gets the number of cross-validation folds used during training.

### Method
get_num_cv_folds()

### Signature
`get_num_cv_folds()`

### Parameters
None.

### Returns
The number of folds.
```

--------------------------------

### Run unit tests

Source: https://interpret.ml/docs/installation-guide.html

Runs the unit tests using make test.

```bash
cd interpret/scripts

make test
```

--------------------------------

### get_cv_y

Source: https://interpret.ml/docs/python/api/APLRRegressor.html

Gets the validation response values (y) for a specific cross-validation fold.

```APIDOC
## get_cv_y

### Description
Gets the validation response values (y) for a specific cross-validation fold.

### Method
get_cv_y(fold_index)

### Signature
`get_cv_y(fold_index)`

### Parameters
- **fold_index** (int) - Required - The index of the fold.

### Returns
A numpy array containing the validation response values.
```

--------------------------------

### get_cv_sample_weight

Source: https://interpret.ml/docs/python/api/APLRRegressor.html

Gets the validation sample weights for a specific cross-validation fold.

```APIDOC
## get_cv_sample_weight

### Description
Gets the validation sample weights for a specific cross-validation fold.

### Method
get_cv_sample_weight(fold_index)

### Signature
`get_cv_sample_weight(fold_index)`

### Parameters
- **fold_index** (int) - Required - The index of the fold.

### Returns
A numpy array containing the validation sample weights.
```

--------------------------------

### get_cv_validation_indexes

Source: https://interpret.ml/docs/python/api/APLRRegressor.html

Gets the original indexes of the validation observations for a specific cross-validation fold.

```APIDOC
## get_cv_validation_indexes

### Description
Gets the original indexes of the validation observations for a specific cross-validation fold.

### Method
get_cv_validation_indexes(fold_index)

### Signature
`get_cv_validation_indexes(fold_index)`

### Parameters
- **fold_index** (int) - Required - The index of the fold.

### Returns
A numpy array containing the original indexes.
```

--------------------------------

### Show local explanations for first 5 test samples

Source: https://interpret.ml/docs/ebm.html

Displays local explanations for the first five test samples using interpret.show.

```python
show(ebm.explain_local(X_test[:5], y_test[:5]), 0)
```

--------------------------------

### Load and prepare the Adult dataset

Source: https://interpret.ml/docs/python/examples/custom-interactions.html

Imports required libraries, loads the Adult dataset from the UCI repository, assigns column names, and creates train/test splits with a fixed random seed.

```python
import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from interpret.glassbox import ExplainableBoostingClassifier

df = pd.read_csv(
    "https://archive.ics.uci.edu/ml/machine-learning-databases/adult/adult.data",
    header=None)
df.columns = [
    "Age", "WorkClass", "fnlwgt", "Education", "EducationNum",
    "MaritalStatus", "Occupation", "Relationship", "Race", "Gender",
    "CapitalGain", "CapitalLoss", "HoursPerWeek", "NativeCountry", "Income"
]
X = df.iloc[:, :-1]
y = df.iloc[:, -1]

seed = 42
np.random.seed(seed)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20, random_state=seed)
```

--------------------------------

### AUC Output from Logistic Regression

Source: https://interpret.ml/docs/lr.html

The printed AUC value from the logistic regression training example, showing 0.998.

```text
AUC: 0.998
```
Tokens:
Raw