#2451 [Feature] Type Transformer for MatPlotLib Issue created by smadarab Describe the bug Type translator for matplotlib chart image doesn't exist. When i try to return a matplotlib image from task to workflow, the rurned datatype is saved as figure in local location but not into s3 bucket. Expected behavior output chart returned from task created using matplotlib should be saved into a output variable and then iunto s3 bucket. Additional context to reproduce from copyreg import pickle import pandas as pd from sklearn.datasets import load_digits from sklearn.linear_model import LogisticRegression import requests import io import numpy as np from sklearn.pipeline import Pipeline, FeatureUnion from sklearn.preprocessing import FunctionTransformer, MinMaxScaler from sklearn.impute import SimpleImputer imputer_num = SimpleImputer(missing_values=np.nan, strategy='mean') from sklearn.model_selection import train_test_split from sklearn.impute import SimpleImputer imputer_cat = SimpleImputer(strategy="most_frequent") from sklearn.ensemble import AdaBoostClassifier from sklearn.metrics import roc_auc_score from flytekit import task, workflow from joblib import dump from sklearn.preprocessing import OneHotEncoder from typing import Tuple from matplotlib import pyplot import pickle hi=None scale=None from sklearn.preprocessing import MinMaxScaler @task def get_dataset() -> pd.DataFrame: #return load_digits(as_frame=True).frame

#url = "<https://github.com/smadarab/flytelab/raw/main/census.csv>" # Make sure the url is the raw version of the file on GitHub
url="<https://github.com/smadarab/flytelab/raw/main/census> 2.csv"
download = requests.get(url).content
df = pd.read_csv(io.StringIO(download.decode('utf-8')),sep=',')
print("df is created",df.columns)
#df.dropna(inplace=True)
#df = df.reset_index()
print(df.columns)
return(df)

@task def plot_dataset(df:pd.DataFrame) -> pyplot: x = df['age'].value_counts().index y = df['age'].value_counts().values fig = pyplot.figure(figsize =(10, 7)) # Horizontal Bar Plot pyplot.bar(x,y) # Show Plot #pyplot.show() return(fig) #@task #def clean_dataset() -> pd.DataFrame: pass #@task #def transform_cat() -> pd.DataFrame: pass #@task #def transform_num() -> pd.DataFrame: pass @task def train_model(train: pd.DataFrame) -> Tuple[AdaBoostClassifier,OneHotEncoder,MinMaxScaler]: num_cols = ['age', 'education-num', 'capital-gain', 'capital-loss', 'hours-per-week'] cat_cols = ['workclass', 'marital-status', 'occupation', 'relationship', 'race', 'sex', 'native-country'] log_transform_cols = ['capital-loss', 'capital-gain'] def get_cat_cols(X): return X[cat_cols] def get_num_cols(X): return X[num_cols] def get_log_transform_cols(X): return X[log_transform_cols] def get_dummies(X): print('\n \n',type(X)) return pd.get_dummies(pd.DataFrame(X)) def cat_imputer(X): print(X.shape) return(imputer_cat.fit_transform(X)) #return X.apply(lambda col: imputer_cat.fit_transform(col)) def one_hot_encode(X): print("one hot encode") ohe = OneHotEncoder(handle_unknown = 'ignore') #print("DF:: ") ohe.fit(pd.DataFrame(X)) global hi hi=ohe #dump(ohe, 'onehot.joblib') return ohe.transform(pd.DataFrame(X)).toarray() def min_max_scaling(X): scaler = MinMaxScaler() scaler.fit(X) global scale scale=scaler return scaler.transform(pd.DataFrame(X)).tolist() log_transform_pipeline = Pipeline([ ('get_log_transform_cols', FunctionTransformer(get_log_transform_cols, validate=False)), ('imputer', SimpleImputer(strategy='mean')), ('log_transform', FunctionTransformer(np.log1p)) ])

num_cols_pipeline = Pipeline([
('get_num_cols', FunctionTransformer(get_num_cols, validate=False)),
('imputer', SimpleImputer(strategy='mean')),
('min_max_scaler', FunctionTransformer(min_max_scaling, validate=False))
])

cat_cols_pipeline = Pipeline([
('get_cat_cols', FunctionTransformer(get_cat_cols, validate=False)),
('imputer', SimpleImputer(strategy="most_frequent")),
#    ('get_dummies', FunctionTransformer(get_dummies, validate=False))
('one_hot_encode', FunctionTransformer(one_hot_encode, validate=False))
])       

steps_ = FeatureUnion([
('log_transform', log_transform_pipeline),
('num_cols', num_cols_pipeline),
('cat_cols', cat_cols_pipeline)
])
full_pipeline = Pipeline([('steps_', steps_)])
y = train['income'].map({'<=50K': 0, '>50K': 1})
X = full_pipeline.fit_transform(train)
model = AdaBoostClassifier(n_estimators=300)
X_train, X_test, y_train, y_test = train_test_split(X, y,shuffle=False)
model = model.fit(X_train, y_train) 
return model,hi,scale

#@task #def test_model() -> pd.DataFrame: pass #@workflow #def main1() -> pd.DataFrame: return get_dataset() @workflow def main() -> pyplot: return plot_dataset(df=get_dataset()) #@workflow #def main() -> Tuple[AdaBoostClassifier,OneHotEncoder,MinMaxScaler]: return train_model(train=get_dataset()) if name == "*main*": print(f"trained model: {main()}") Screenshots No response Are you sure this issue hasn't been raised already? ☑︎ Yes Have you read the Code of Conduct? ☑︎ Yes flyteorg/flyte