def test_preparedata_runs_without_crashing(self):
df_test = pd.read_csv(
'restapi/machine_learning/testfiles/df_minimal.csv')
target_variable = 'G3_class'
factor_list = ['failures', 'C(sex)']
y_expected = np.array([0., 0., 1., 0., 1., 1., 0., 1., 1.])
y, X = preparedata(df_test, target_variable, factor_list)
np.testing.assert_array_equal(y, y_expected)
def retrain_model(request):
if request.method == 'POST':
data = json.loads(request.body.decode('utf-8'))
if "model_id" not in data or data["model_id"] is None:
model_id = 1
else:
model_id = data["model_id"]
mlmodel = MlModel.objects.get(pk=model_id)
dataset = mlmodel.dataset_id
target_variable = mlmodel.target_variable
dataFilePath = os.path.join(settings.MEDIA_ROOT, dataset.file.name)
numeric_columns = get_numeric_columns(model_id)
factor_list_wo_categories = get_factor_list_from_file(
dataFilePath, target_variable, numeric_columns)
model, model_description = retrain(factor_list_wo_categories,
target_variable,
data["factors"],
dataFile=dataFilePath)
# not enforced here: error if two factors are balanced
protected_attr = None
for factor in data["factors"]:
if factor["is_balanced"]:
protected_attr = factor["name"]
df_data = pd.read_csv(dataFilePath)
y, X = preparedata(df_data, target_variable, factor_list_wo_categories)
X = drop_disabled_factors(X, data["factors"])
if protected_attr is not None:
if protected_attr not in X:
raise ValueError(
'missing protected_attr in X, it may have been disabled')
thresholds = get_fair_thresholds(model, X, y, protected_attr)
model_description["negative_threshold"] = thresholds[0]
model_description["positive_threshold"] = thresholds[1]
print(thresholds)
accuracy, confusion_matrices = test_logreg_model(
model, X, y, thresholds, protected_attr)
else:
accuracy, confusion_matrices = test_logreg_model(model, X, y)
model_description['accuracy'] = accuracy
model_description['confusion_matrices'] = confusion_matrices
return Response(model_description, status=status.HTTP_200_OK)
else:
return Response('HTTP_400_BAD_REQUEST',
status=status.HTTP_400_BAD_REQUEST)
def test_model(request):
if request.method == 'POST':
data = json.loads(request.body.decode('utf-8'))
model_id = data["model_id"]
mlmodel = MlModel.objects.get(pk=int(model_id))
target_variable = mlmodel.target_variable
dataset = mlmodel.dataset_id
target_variable = mlmodel.target_variable
dataFilePath = os.path.join(settings.MEDIA_ROOT, dataset.file.name)
protected_attr = None
for factor in data["factors"]:
if factor["is_balanced"]:
protected_attr = factor["name"]
thresholds = None
if protected_attr is not None:
if ("positive_threshold" in data and "negative_threshold" in data
and data["negative_threshold"] is not None
and data["positive_threshold"] is not None):
thresholds = (data["negative_threshold"],
data["positive_threshold"])
print(thresholds)
else:
return Response('Retrain needed, thresholds missing',
status=status.HTTP_400_BAD_REQUEST)
numeric_columns = get_numeric_columns(model_id)
factor_list_wo_categories = get_factor_list_from_file(
dataFilePath, target_variable, numeric_columns)
df_data = pd.read_csv(dataFilePath)
y, X = preparedata(df_data, target_variable, factor_list_wo_categories)
X = drop_disabled_factors(X, data["factors"])
for factor in data["factors"]:
if isinstance(factor["weight"], str):
raise ValueError(
'factor weights cannot be strings, but are for ' +
factor["name"])
model = build_model_from_factors(data["factors"], data["intercept"], y,
X)
accuracy, confusion_matrices = test_logreg_model(
model, X, y, thresholds, protected_attr)
res = {'accuracy': accuracy, 'confusion_matrices': confusion_matrices}
return Response(res, status=status.HTTP_200_OK)
return Response('HTTP_400_BAD_REQUEST', status=status.HTTP_400_BAD_REQUEST)
def retrain(factor_list_wo_categories, target_variable, factors, dataFile):
df_math = pd.read_csv(dataFile)
y, X = preparedata(df_math, target_variable, factor_list_wo_categories)
for factor in factors:
if not factor["is_enabled"]:
X = X.drop(factor["name"], axis=1)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.25,
random_state=42)
model = LogisticRegression()
model.fit(X_train, y_train)
# enabling/disabling factors
for factor in factors:
col_list = X.dtypes.index.tolist()
factor_name = factor['name']
if factor_name in col_list:
col_num_in_X = col_list.index(factor_name)
factor["weight"] = model.coef_[0][col_num_in_X]
return model, {'factors': factors}
def new_model_with_factor_creation(request):
# expects JSON with name, description, and dataset_id
if request.method != 'POST':
return Response('HTTP_400_BAD_REQUEST',
status=status.HTTP_400_BAD_REQUEST)
data = json.loads(request.body.decode('utf-8'))
# Save new model
dataset_id = data['dataset_id']
non_categorical_columns = data['non_categorical_columns']
target_variable = data['target_variable']
target_var_alias = data['target_var_alias']
newModel = MlModel(name=data['name'],
description=data['description'],
dataset_id=DataSet.objects.get(pk=dataset_id),
modified=timezone.now(),
non_categorical_columns=non_categorical_columns,
target_variable=target_variable,
target_var_alias=target_var_alias)
newModel.save()
# Get factor names from file
dataset = DataSet.objects.get(pk=dataset_id)
datafile = dataset.file
dataFilePath = 'datasets/' + datafile.name
numeric_columns = non_categorical_columns.split(',')
factor_list = get_factor_list_from_file(dataFilePath, target_variable,
numeric_columns)
y, X = preparedata(pd.read_csv(dataFilePath), target_variable, factor_list)
factors_to_save = X.columns.values.tolist()
# Save factors
for factor in factors_to_save:
newFactor = Factor(name=factor,
alias=factor,
weight=0,
model_id=newModel)
newFactor.save()
return Response(newModel.id, status=status.HTTP_200_OK)