Scikit Learn IsolationForest: How to Fit Multiple Dataframes With Different Parameters (Not Using GridSearchCV)

Question

I have a five separate pandas dataframes that I've put inside a dictionary. I want to run five separate IsolationForest models in scikit-learn with different sets of parameters for each model. However, I don't want to run each model separtely.

So my question is, how can I run these models and get the decision functions and predictions for all dataframes in on go. My attempt at doing so is below.

# parameters for each of the five models in a list. The index position in the list
# represents a dataset's parameters, from dataset0 through dataset4

n_estimators = [150, 200, 125, 125, 125]
max_samples = [0.70, 0.70, 0.80, 0.70, 0.70]
max_features = [1, 4, 2, 2, 3]
contamination = [0.05, 0.06, 0.05, 0.07, 0.05]

# numeric columns
num_columns = list(subset_features[1:])

# column transformer
num_transformer = Pipeline([('impute', IterativeImputer()), ('scale', StandardScaler())])
ct = ColumnTransformer([('num_pipeline', num_transformer, num_columns)])

clf = Pipeline([('ct', ct),
                ('iforest', IsolationForest(n_estimators=n_estimators[i],
                                            max_samples=max_samples[i],
                                            max_features=max_features[i],
                                            contamination=contamination[i],
                                            n_jobs=4,
                                            random_state=None))])



clf_res = {}
for i, df in enumerate(dfs.values()):
    print('starting idx:', i)
    clf_res[i] = clf.fit(df)

The issue I have is that it is not iterating through the different sets of parameters as the dataframes change from iteration to iteration. See below:

{0: Pipeline(steps=[('ct',
                  ColumnTransformer(transformers=[('numeric_pipeline',
                                                   Pipeline(steps=[('impute',
                                                                    IterativeImputer()),
                                                                   ('scale',
                                                                    StandardScaler())]),
                                                   ['V1', 'V2', 'V3',
                                                    'V4'])])),
                 ('iforest',
                  IsolationForest(contamination=0.05, max_features=1,
                                  max_samples=0.7, n_estimators=125,
                                  n_jobs=4))]),
 1: Pipeline(steps=[('ct',
                  ColumnTransformer(transformers=[('numeric_pipeline',
                                                   Pipeline(steps=[('impute',
                                                                    IterativeImputer()),
                                                                   ('scale',
                                                                    StandardScaler())]),
                                                   ['V1', 'V2', 'V3',
                                                    'V4'])])),
                 ('iforest',
                  IsolationForest(contamination=0.05, max_features=1,
                                  max_samples=0.7, n_estimators=125,
                                  n_jobs=4))])

So what I want is that the parameters will change as the dataframes changes.

Answer 1

After careful review of my code, and a little bit of googling, I found out what was wrong with my code. I am sharing it here for anyone else who may have similar problems in the future.

The only change I made to my code was to bring clf pipeline into the for loop.

# parameters for each of the five models in a list. The index position in the list
# represents a dataset's parameters, from dataset0 through dataset3

n_estimators = [150, 200, 125, 125, 125]
max_samples = [0.70, 0.70, 0.80, 0.70, 0.70]
max_features = [1, 4, 2, 2, 3]
contamination = [0.05, 0.06, 0.05, 0.07, 0.05]

# numeric columns
num_columns = list(subset_features[1:])

# column transformer
num_transformer = Pipeline([('impute', IterativeImputer()), ('scale', StandardScaler())])
ct = ColumnTransformer([('num_pipeline', num_transformer, num_columns)])

clf_res = {}
for i, df in enumerate(dfs.values()):
    print('starting idx:', i)
    clf = Pipeline([('ct', ct),
                    ('iforest', IsolationForest(n_estimators=n_estimators[i],
                                                max_samples=max_samples[i],
                                                max_features=max_features[i],
                                                contamination=contamination[i],
                                                n_jobs=4,
                                                random_state=None))])
    clf_res[i] = clf.fit(df)

Sample of correct output is below

{0: Pipeline(steps=[('ct',
                  ColumnTransformer(transformers=[('numeric_pipeline',
                                                   Pipeline(steps=[('impute',
                                                                    IterativeImputer()),
                                                                   ('scale',
                                                                    StandardScaler())]),
                                                   ['v1', 'v2', 'v3',
                                                    'v4'])])),
                 ('iforest',
                  IsolationForest(contamination=0.05, max_features=1,
                                  max_samples=0.70,
                                  n_estimators=150, n_jobs=4))]),
 1: Pipeline(steps=[('ct',
                  ColumnTransformer(transformers=[('numeric_pipeline',
                                                   Pipeline(steps=[('impute',
                                                                    IterativeImputer()),
                                                                   ('scale',
                                                                    StandardScaler())]),
                                                   ['v1', 'v2', 'v3',
                                                    'v4'])])),
                 ('iforest',
                  IsolationForest(contamination=0.05, max_features=4,
                                  max_samples=0.7, n_estimators=200,
                                  n_jobs=4))]),