def test_explain_smoke_titanic():
titanic = load_titanic()
titanic_clean = clean(titanic)
sc = SimpleClassifier().fit(titanic_clean, target_col='survived')
explain(sc)
X, y = titanic_clean.drop("survived", axis=1), titanic_clean.survived
ep = EasyPreprocessor()
preprocessed = ep.fit_transform(X)
tree = DecisionTreeClassifier().fit(preprocessed, y)
explain(tree, feature_names=ep.get_feature_names())
pipe = make_pipeline(EasyPreprocessor(), LogisticRegression())
pipe.fit(X, y)
explain(pipe, feature_names=pipe[0].get_feature_names())
def test_explain_titanic_val(model):
# add multi-class
# add regression
titanic = load_titanic()
titanic_clean = clean(titanic)
X, y = titanic_clean.drop("survived", axis=1), titanic_clean.survived
X_train, X_val, y_train, y_val = train_test_split(X, y, stratify=y,
random_state=42)
pipe = make_pipeline(EasyPreprocessor(), model)
pipe.fit(X_train, y_train)
# without validation set
explain(pipe, feature_names=X.columns)
# with validation set
explain(pipe, X_val, y_val, feature_names=X.columns)
def preprocess_data(df, sample=20000):
data = df.sample(n=min(sample, len(df)), random_state=42)
data_clean, data_types = dabl.clean(data, return_types=True, verbose=3)
return data_clean, data_types
def run(csv_metadata, list_files, output_folder):
csv_id = csv_metadata[0]
csv_metadata = csv_metadata[1]
tqdm.write(f"\nTreating {csv_id} file")
#Find the full path for the csv_id
mask = [csv_id in str(f) for f in list_files]
csv_file_path = list_files[mask.index(True)]
dabl_analysis_path = (output_folder /
(csv_id + '_dabl')).with_suffix('.csv')
"""
if dabl_analysis_path.exists():
tqdm.write(f"File {csv_id} already analyzed: {dabl_analysis_path} already exists")
return dabl_analysis_path"""
result_list = []
if csv_metadata and len(csv_metadata) > 1:
encoding = csv_metadata["encoding"]
sep = csv_metadata["separator"]
else:
encoding = "latin-1" # because why not
sep = ";"
csv_detective_columns = []
if "columns" in csv_metadata:
# keep columns that are not boolean
csv_detective_columns = [
k.strip('"') for k, v in csv_metadata['columns'].items()
if "booleen" not in v
]
try:
data: pd.DataFrame = pd.read_csv(csv_file_path.as_posix(),
encoding=encoding,
sep=sep,
error_bad_lines=False)
# remove csv_detective columns
#data = data.drop(csv_detective_columns, axis=1)
# TODO change this as now the columns are not in the same order
data_clean, data_types = dabl.clean(data, return_types=True, verbose=3)
# dabl.detect_types(data)
money_variables = csv_metadata['columns']['money']
for target_col in money_variables:
try:
data_clean_no_nan = data_clean[data_clean[target_col].notna()]
if len(data_clean_no_nan
) < 100: # less than 100 examples is too few examples
continue
print(f"Building models with target variable: {target_col}")
sc = dabl.SimpleRegressor(random_state=42).fit(
data_clean_no_nan, target_col=target_col)
features_names = sc.est_.steps[0][1].get_feature_names()
inner_dict = {
"csv_id": csv_id,
"task": "regression",
"algorithm": sc.current_best_.name,
"target_col": target_col,
"nb_features": len(features_names),
"features_names": "|".join(features_names),
"nb_classes": len(data[target_col].unique()),
"nb_lines": data_clean_no_nan.shape[0],
"date": today,
}
inner_dict.update(sc.current_best_.to_dict())
inner_dict.update({
"avg_scores":
np.mean(list(sc.current_best_.to_dict().values()))
})
result_list.append(inner_dict)
except Exception as e:
tqdm.write(
f"Could not analyze file {csv_id} with target col {target_col}. Error {str(e)}"
)
except Exception as e:
tqdm.write(f"Could not analyze file {csv_id}. Error: {e}")
return None
if not result_list:
return
result_df = pd.DataFrame(result_list)
with open(dabl_analysis_path, "w") as filo:
result_df.to_csv(filo, header=True, index=False)
return dabl_analysis_path
bdf.save_within_48.value_counts()
# ## Extract and do DABL
# In[ ]:
import dabl
# In[ ]:
feature_df = bdf[final_feature_list]
# In[ ]:
dabl_data = dabl.clean(feature_df)
# In[ ]:
dabl.plot(dabl_data, target_col='save_within_48')
# In[ ]:
X = dabl_data.drop("save_within_48", axis=1)
Y = dabl_data.save_within_48
# In[ ]:
preprocessor = dabl.EasyPreprocessor()
X_trans = preprocessor.fit_transform(X)
