IsoTree to TreeLite¶

This is a short example of converting an Isolation Forest model generated through the isotree library to treelite format, which can then be passed to the tl2cgen library to compile these trees to a standalone runtime library which is oftentimes faster ar making predictions.

Getting some medium-size data from scikit-learn to fit a model¶

In [1]:

from sklearn.datasets import fetch_california_housing

X, y = fetch_california_housing(return_X_y=True)
print(X.shape)

(20640, 8)

Fitting an isolation forest model through isotree¶

Note: only models that use ndim=1 can be exported to treelite format.

In [2]:

from isotree import IsolationForest

iso = IsolationForest(ndim=1, ntrees=100, sample_size=256,
                      missing_action="impute", max_depth=8)
iso.fit(X)

### Now convert
treelite_model = iso.to_treelite()

### OPTIONAL: add annotations for better branch prediction
import tl2cgen
tl2cgen.annotate_branch(
    model=treelite_model,
    dmat=tl2cgen.DMatrix(X),
    verbose=False,
    path="iso_branches_annotation.json"
)

Compiling the treelite model¶

These models need to be compiled into a shared library in order to be used:

In [3]:

%%capture
import tl2cgen
import multiprocessing

tl2cgen.export_lib(
    model=treelite_model,
    toolchain="clang",
    libpath='./predictor.so',
    params={
        "parallel_comp": multiprocessing.cpu_count(),
        "annotate_in": "iso_branches_annotation.json"
    }
)
treelite_predictor = tl2cgen.Predictor("predictor.so")

Now verify that they make the same predictions:

In [4]:

iso.predict(X[:10])

Out[4]:

array([0.47006444, 0.47770081, 0.4910637 , 0.42605826, 0.41548625,
       0.41730139, 0.41699421, 0.43228664, 0.40877799, 0.41800632])

In [5]:

treelite_predictor.predict(tl2cgen.DMatrix(X[:10]))

Out[5]:

array([[0.47006445],
       [0.47770081],
       [0.4910637 ],
       [0.42605827],
       [0.41548626],
       [0.41730139],
       [0.41699421],
       [0.43228664],
       [0.40877799],
       [0.41800632]])

Note: some small disagreement between the two is expected due to loss of precision when converting. See the documentation in isotree for more details.

Comparing prediction times¶

In [6]:

%%timeit
import joblib
### see docs for 'IsolationForest.predict' about this part
iso.set_params(nthreads=joblib.cpu_count(only_physical_cores=True))
iso.predict(X)

20.4 ms ± 423 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

In [7]:

%%timeit
treelite_predictor.predict(tl2cgen.DMatrix(X))

4.14 ms ± 620 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)