def train_gefs_model():
print("Preparing df for GeFs Random forest model")
df, ncat = get_fraud_data(df) # Preprocess the df
# ncat is the number of categories of each variable in the df
X_train, X_test, y_train, y_test, data_train, data_test = gefs_train_test_split(
df, ncat)
rf = RandomForest(n_estimators=30, ncat=ncat) # Train a Random Forest
print('Starting the GeFs Random Forest Training')
rf.fit(X_train, y_train)
print('Converting Random Forest to GeF')
gef = rf.topc() # Convert to a GeF
## Classification is performed either by averaging the prediction of each tree (`classify_avg` method)
# or by defining a mixture over them (`classify` method).
print('Making predictions on test df')
y_pred_avg = gef.classify_avg(X_test, classcol=df.shape[1] - 1)
y_pred_mixture = gef.classify(X_test, classcol=df.shape[1] - 1)
_, y_prob = gef.classify(X_test,
classcol=df.shape[1] - 1,
return_prob=True)
y_prob = np.max(y_prob, axis=1)
from sklearn import metrics
score = metrics.roc_auc_score(y_test, y_prob)
print(f"Test score for GeFs Model: {score}")
def train_gefs_model():
print("Preparing data for GeFs Random forest model")
data, ncat = get_fraud_data(df) # Preprocess the data
# ncat is the number of categories of each variable in the data
X_train, X_test, y_train, y_test, data_train, data_test = gefs_train_test_split(
data, ncat)
rf = RandomForest(n_estimators=30, ncat=ncat) # Train a Random Forest
print('Starting the GeFs Random Forest Training')
rf.fit(X_train, y_train)
print('Converting Random Forest to GeF')
gef = rf.topc() # Convert to a GeF
## Classification is performed either by averaging the prediction of each tree (`classify_avg` method)
# or by defining a mixture over them (`classify` method).
print('Making predictions on test data')
y_pred_avg = gef.classify_avg(X_test, classcol=data.shape[1] - 1)
y_pred_mixture = gef.classify(X_test, classcol=data.shape[1] - 1)
from sklearn import metrics
score = metrics.roc_auc_score(y_test, y_pred_avg)
print(f"Test score for GeFs Model: {score}")
### Computing Robustness Values
## Robustness values can be computed with the `compute_rob_class` function.
from gefs import compute_rob_class
pred, rob = compute_rob_class(gef.root, X_test, data.shape[1] - 1,
int(ncat[-1]))
def fit(data_pars=None, compute_pars=None, out_pars=None, **kw):
"""
"""
global model, session
session = None # Session type for compute
Xtrain, ytrain, Xtest, ytest = get_dataset(data_pars, task_type="train")
if VERBOSE: log(Xtrain.shape, model.model)
if model.ncat is None:
log("#!IMPORTANT This indicates that the preprocessing pipeline was not adapted to GEFS! and we need to calculate ncat"
)
cont_cols = data_pars['cols_input_type'].get(
"colnum") # continous, float column is this correct?
temp_train = pd.concat([Xtrain, ytrain], axis=1)
temp_test = pd.concat([Xtest, ytest], axis=1)
df = pd.concat([temp_train, temp_test], ignore_index=True, sort=False)
model.ncat = pd_colcat_get_catcount(
df.values,
classcol=-1,
continuous_ids=[df.columns.get_loc(c) for c in cont_cols])
model.model = RandomForest(model.n_estimators, ncat=model.ncat)
model.model.fit(Xtrain, ytrain)
model.model = model.model.topc() # Convert to a GeF
def fit(data_pars=None, compute_pars=None, out_pars=None, **kw):
"""
"""
global model, session
session = None # Session type for compute
Xtrain, ytrain, Xtest, ytest = get_dataset(data_pars, task_type="train")
log(Xtrain.shape, model.model)
if model.ncat is None:
log("#!IMPORTANT This indicates that the preprocessing pipeline was not adapted to GEFS! and we need to calculate ncat"
)
cont_cols = data_pars['cols_input_type'].get(
"colnum") # continous, float column is this correct?
temp_train = pd.concat([Xtrain, ytrain], axis=1)
temp_test = pd.concat([Xtest, ytest], axis=1)
df = pd.concat([temp_train, temp_test], ignore_index=True, sort=False)
model.ncat = pd_colcat_get_catcount(
df,
# categ cols
colcat=data_pars["cols_input_type"]["colcat"],
# target col index
classcol=-1,
# num cols indices
continuous_ids=[df.columns.get_loc(c) for c in cont_cols])
ncat = np.array(list(model.ncat.values()))
# In case of warnings make sure ncat is consistent
# check this issue : https://github.com/AlCorreia/GeFs/issues/6
"""
def __init__(self, n_estimators=100, imp_measure='gini', min_samples_split=2,
min_samples_leaf=1, max_features=None, bootstrap=True,
ncat=None, max_depth=1e6, surrogate=False):
"""
model.model = RandomForest(
n_estimators=model.n_estimators,
ncat=ncat,
)
# Remove the target col
X = Xtrain.iloc[:, :-1]
# y should be 1-dim
model.model.fit(X.values, ytrain.values.reshape(-1))
# Make sure ncat is consistent, otherwise model.topc()
# will throw all kind of numba errors
# check this issue : https://github.com/AlCorreia/GeFs/issues/5
model.model = model.model.topc() # Convert to a GeF
def __init__(self, model_pars=None, data_pars=None, compute_pars=None):
self.model_pars, self.compute_pars, self.data_pars = model_pars, compute_pars, data_pars
if model_pars is None:
self.model = None
else:
self.n_estimators = model_pars.get('n_estimators', 100)
self.ncat = model_pars.get(
'ncat', None
) # Number of categories of each variable This is an ndarray
if self.ncat is None:
self.model = None # In order to create an instance of the model we need to calculate the ncat mentioned above on our dataset
log('ncat is not define')
else:
self.model = RandomForest(n_estimators=self.n_estimators,
ncat=self.ncat)
if VERBOSE: log(None, self.model)
def __init__(self, model_pars=None, data_pars=None, compute_pars=None):
self.model_pars, self.compute_pars, self.data_pars = model_pars, compute_pars, data_pars
if model_pars is None:
self.model = None
else:
self.n_estimators = model_pars.get('n_estimators', 100)
self.ncat = model_pars.get(
'ncat', None
) # Number of categories of each variable This is an ndarray
if self.ncat is None:
self.model = None # In order to create an instance of the model we need to calculate the ncat mentioned above on our dataset
log('ncat is not define')
else:
"""
def __init__(self, n_estimators=100, imp_measure='gini', min_samples_split=2,
min_samples_leaf=1, max_features=None, bootstrap=True,
ncat=None, max_depth=1e6, surrogate=False):
"""
self.model = RandomForest(n_estimators=self.n_estimators,
ncat=self.ncat)
log(None, self.model)
data.insert(
len(data.columns) - 1, 'is_attributed', data.pop('is_attributed'))
data.loc[:, cat_cols] = get_dummies(data[cat_cols])
ncat = learncats(
data.values,
classcol=-1,
continuous_ids=[data.columns.get_loc(c) for c in cont_cols])
return data.values.astype(float), ncat
print("Preparing data for GeFs Random forest model")
data, ncat = get_fraud_data(df) # Preprocess the data
# ncat is the number of categories of each variable in the data
X_train, X_test, y_train, y_test, data_train, data_test = gefs_train_test_split(
data, ncat)
rf = RandomForest(n_estimators=30, ncat=ncat) # Train a Random Forest
print('Starting the GeFs Random Forest Training')
rf.fit(X_train, y_train)
print('Converting Random Forest to GeF')
gef = rf.topc() # Convert to a GeF
from sklearn.ensemble import RandomForestClassifier as rfsk
rfsk.fit(X_train, y_train)
y_pred_avg_sk = rfsk.predict(X_test)
## Classification is performed either by averaging the prediction of each tree (`classify_avg` method)
# or by defining a mixture over them (`classify` method).
print('Making predictions on test data')
y_pred_avg = gef.classify_avg(X_test, classcol=data.shape[1] - 1)
y_pred_mixture = gef.classify(X_test, classcol=data.shape[1] - 1)