def handle(self, *args, **kwargs):
# get model
TARGET_MODEL = 59
job = Job.objects.filter(pk=TARGET_MODEL)[0]
model = joblib.load(job.predictive_model.model_path)[0]
# load data
training_df, test_df = get_encoded_logs(job)
training_df['label'] = training_df['label'].astype(bool).astype(int)
columns = list(training_df.columns.values)
features = list(
training_df.drop(['trace_id', 'label'], 1).columns.values)
feature = 'Age_1'
feature_grids, percentile_info = _get_grids(
feature_values=training_df[feature].values,
num_grid_points=10,
grid_type=None,
percentile_range='percentile',
grid_range=None)
custom_grids = []
indexs = []
for x in range(int(feature_grids.min()), int(feature_grids.max() - 1)):
custom_grids.append(x)
print(features)
fig, axes, summary_df = info_plots.target_plot(
df=training_df,
feature=feature,
feature_name='feature value',
cust_grid_points=custom_grids,
target='label',
show_percentile=False)
fig.savefig('ice_plot_train_1_3_CType.png')
lists = list(training_df[feature].values)
for x in range(int(feature_grids.min()), int(feature_grids.max() - 1)):
indexs.append(lists.index(x))
encoder = retrieve_proper_encoder(job)
encoder.decode(training_df, job.encoding)
values = training_df[feature].values
training_df
lst = []
print(summary_df)
if job.encoding.value_encoding != ValueEncodings.BOOLEAN.value:
for x in range(len(indexs) - 1):
lst.append({
'value': values[indexs[x]],
'label': summary_df['label'][x],
'count': summary_df['count'][x],
})
else:
for x in range(summary_df.shape[0]):
lst.append({
'value': summary_df['display_column'][x],
'label': summary_df['label'][x],
'count': summary_df['count'][x],
})
print(lst)
def test_grids_grid_range_equal(self, titanic_data):
# grid_range=(0, 100), grid_type='equal'
feature_grids, percentile_info = _get_grids(
feature_values=titanic_data['Fare'].values, num_grid_points=10, grid_type='equal',
percentile_range=None, grid_range=(0, 100))
expected_feature_grids = np.array([0.0, 44.44444444, 77.77777777])
expected_percentile_info = []
assert_array_almost_equal(feature_grids[[0, 4, 7]], expected_feature_grids, decimal=8)
assert percentile_info == expected_percentile_info
def test_grids_grid_range_percentile(self, titanic_data):
# grid_range=(0, 100), grid_type='percentile'
feature_grids, percentile_info = _get_grids(
feature_values=titanic_data['Fare'].values, num_grid_points=10, grid_type='percentile',
percentile_range=None, grid_range=(0, 100))
expected_feature_grids = np.array([0.0, 13.0, 35.11111111])
expected_percentile_info = np.array(['(0.0)', '(44.44)', '(77.78)'])
assert_array_almost_equal(feature_grids[[0, 4, 7]], expected_feature_grids, decimal=8)
assert_array_equal(percentile_info[[0, 4, 7]], expected_percentile_info)
def test_grids_percentile_range(self, titanic_data):
# percentile_range=(5, 95)
feature_grids, percentile_info = _get_grids(
feature_values=titanic_data['Fare'].values, num_grid_points=10, grid_type='percentile',
percentile_range=(5, 95), grid_range=None)
expected_feature_grids = np.array([7.225, 13.0, 31.0])
expected_percentile_info = np.array(['(5.0)', '(45.0)', '(75.0)'])
assert_array_equal(feature_grids[[0, 4, 7]], expected_feature_grids)
assert_array_equal(percentile_info[[0, 4, 7]], expected_percentile_info)
def test_grids_15(self, titanic_data):
# num_grid_points=15
feature_grids, percentile_info = _get_grids(
feature_values=titanic_data['Fare'].values, num_grid_points=15, grid_type='percentile',
percentile_range=None, grid_range=None)
expected_feature_grids = np.array([0.0, 8.05, 14.4542, 37.0042])
expected_percentile_info = np.array(['(0.0)', '(28.57)', '(50.0)', '(78.57)'])
assert_array_equal(feature_grids[[0, 4, 7, 11]], expected_feature_grids)
assert_array_equal(percentile_info[[0, 4, 7, 11]], expected_percentile_info)
def explain(ice_exp: Explanation, training_df, test_df, explanation_target,
prefix_target):
job = ice_exp.job
training_df = training_df.drop(['trace_id'], 1)
if job.encoding.value_encoding == ValueEncodings.BOOLEAN.value:
training_df['label'] = training_df['label'].astype(bool).astype(
int) + 1
feature_grids, percentile_info = _get_grids(
feature_values=training_df[explanation_target].values,
num_grid_points=10,
grid_type=None,
percentile_range='percentile',
grid_range=None)
custom_grids = [
x
for x in range(int(feature_grids.min()), int(feature_grids.max() - 1))
]
fig, axes, summary_df = info_plots.target_plot(
df=training_df,
feature=explanation_target,
feature_name='feature value',
cust_grid_points=custom_grids,
target='label',
show_percentile=False)
lists = list(training_df[explanation_target].values)
indexs = [
lists.index(x)
for x in range(int(feature_grids.min()), int(feature_grids.max() - 1))
]
encoder = retrieve_proper_encoder(job)
encoder.decode(training_df, job.encoding)
values = training_df[explanation_target].values
lst = []
if job.encoding.value_encoding != ValueEncodings.BOOLEAN.value:
for x in range(len(indexs) - 1):
lst.append({
'value': values[indexs[x]],
'label': summary_df['label'][x],
'count': int(summary_df['count'][x]),
})
else:
for x in range(summary_df.shape[0]):
lst.append({
'value': summary_df['display_column'][x],
'label': summary_df['label'][x],
'count': int(summary_df['count'][x]),
})
return lst
def test_grids_grid_type(self, titanic_data):
# grid_type='equal'
feature_grids, percentile_info = _get_grids(
feature_values=titanic_data["Fare"].values,
num_grid_points=10,
grid_type="equal",
percentile_range=None,
grid_range=None,
)
expected_feature_grids = np.array([0.0, 227.70186666, 398.47826666])
expected_percentile_info = []
assert_array_almost_equal(feature_grids[[0, 4, 7]],
expected_feature_grids,
decimal=8)
assert percentile_info == expected_percentile_info
def test_grids_default(self, titanic_data):
# default setting
feature_grids, percentile_info = _get_grids(
feature_values=titanic_data["Fare"].values,
num_grid_points=10,
grid_type="percentile",
percentile_range=None,
grid_range=None,
)
expected_feature_grids = np.array([0.0, 13.0, 35.11111111])
expected_percentile_info = np.array(["(0.0)", "(44.44)", "(77.78)"])
assert_array_almost_equal(feature_grids[[0, 4, 7]],
expected_feature_grids,
decimal=8)
assert_array_equal(percentile_info[[0, 4, 7]],
expected_percentile_info)
