ISEbd-31_Baygulov_A.A._MAI/backend/service.py

60 lines
2.1 KiB
Python
Raw Permalink Normal View History

2024-11-22 19:48:48 +04:00
import io
import os
import pathlib
from typing import BinaryIO, Dict, List
import pandas as pd
from matplotlib.figure import Figure
from werkzeug.datastructures import FileStorage
from werkzeug.utils import secure_filename
class Service:
def __init__(self, dataset_path: str | None) -> None:
if dataset_path is None:
raise Exception("Dataset path is not defined")
self.__path: str = dataset_path
def __get_dataset(self, filename: str) -> pd.DataFrame:
full_file_name = os.path.join(self.__path, secure_filename(filename))
return pd.read_csv(full_file_name)
def upload_dataset(self, file: FileStorage) -> str:
if file.filename is None:
raise Exception("Dataset upload error")
file_name: str = file.filename
full_file_name = os.path.join(self.__path, secure_filename(file_name))
file.save(full_file_name)
return file_name
def get_all_datasets(self) -> List[str]:
return [file.name for file in pathlib.Path(self.__path).glob("*.csv")]
def get_dataset_info(self, filename) -> List[Dict]:
dataset = self.__get_dataset(filename)
dataset_info = []
for column in dataset.columns:
items = dataset[column].astype(str)
column_info = {
"name": column,
"datatype": dataset.dtypes[column],
"items": items,
}
dataset_info.append(column_info)
return dataset_info
def get_column_info(self, filename, column) -> Dict:
dataset = self.__get_dataset(filename)
datatype = dataset.dtypes[column]
items = sorted(dataset[column].astype(str).unique())
return {"datatype": datatype, "items": items}
def get_hist(self, filename, column) -> BinaryIO:
dataset = self.__get_dataset(filename)
bytes = io.BytesIO()
plot: Figure | None = dataset.plot.hist(column=[column], bins=80).get_figure()
if plot is None:
raise Exception("Can't create hist plot")
plot.savefig(bytes, dpi=300, format="png")
return bytes