path: root/src/argaze/DataFeatures.py

"""Miscellaneous data features."""

__author__ = "Théo de la Hogue"
__credits__ = []
__copyright__ = "Copyright 2023, Ecole Nationale de l'Aviation Civile (ENAC)"
__license__ = "BSD"

from typing import TypeVar, Tuple, Any
import os
import sys
import traceback
import importlib
from inspect import getmembers, getmodule
import collections
import json
import ast
import bisect
import threading
import math
import time

import pandas
import numpy
import matplotlib.pyplot as mpyplot
import matplotlib.patches as mpatches
from colorama import Style, Fore

TimeStampType = TypeVar('TimeStamp', int, float)
"""Type definition for timestamp as integer or float values."""

TimestampedObjectType = TypeVar('TimestampedObject', bound="TimestampedObject")
# Type definition for type annotation convenience

TimestampedObjectsListType = TypeVar('TimestampedObjectsList', bound="TimestampedObjectsList")
# Type definition for type annotation convenience

def module_path(obj) -> str:
    """
    Get object module path.

    Returns:
        module path
    """
    return obj.__class__.__module__

def properties(cls) -> list:
    """get class properties name."""

    properties = [name for name, item in cls.__dict__.items() if isinstance(item, property)]
    
    for base in cls.__bases__:

        for name, item in base.__dict__.items(): 

            if isinstance(item, property):

                properties.append(name)

    return properties

def as_dict(obj, filter: bool=True) -> dict:
    """Export object as dictionary.

    Parameters:
        filter: remove None attribute values.
    """
    _dict = {}

    for p in properties(obj.__class__):

        v = getattr(obj, p)

        if not filter or v is not None:

            _dict[p] = v

    return _dict

class JsonEncoder(json.JSONEncoder):
    """Specific ArGaze JSON Encoder."""

    def default(self, obj):
        """default implementation to serialize object."""

        # numpy cases
        if isinstance(obj, numpy.integer):
            return int(obj)

        elif isinstance(obj, numpy.floating):
            return float(obj)

        elif isinstance(obj, numpy.ndarray):
            return obj.tolist()

        # default case
        try:

            return json.JSONEncoder.default(self, obj)

        # class case
        except:

            # ignore attribute starting with _
            public_dict = {}

            for k, v in vars(obj).items():
                
                if not k.startswith('_'):
                    
                    # numpy cases
                    if isinstance(v, numpy.integer):
                        v = int(v)

                    elif isinstance(v, numpy.floating):
                        v = float(v)

                    elif isinstance(v, numpy.ndarray):
                        v = v.tolist()

                    public_dict[k] = v

            return public_dict

class DataDictionary(dict):
    """Enable dot.notation access to dictionary attributes"""

    __getattr__ = dict.get
    __setattr__ = dict.__setitem__
    __delattr__ = dict.__delitem__

class TimestampedObject():
    """Abstract class to enable timestamp management."""

    def __init__(self, timestamp: int|float = math.nan):
        """Initialize TimestampedObject."""
        self._timestamp = timestamp

    def __repr__(self):
        """String representation."""
        return json.dumps(as_dict(self))

    @property
    def timestamp(self) -> int|float:
        """Get object timestamp."""
        return self._timestamp

    @timestamp.setter
    def timestamp(self, timestamp: int|float):
        """Set object timestamp."""
        self._timestamp = timestamp

    def untimestamp(self):
        """Reset object timestamp."""
        self._timestamp = math.nan

    def is_timestamped(self) -> bool:
        """Is the object timestamped?"""
        return not math.isnan(self._timestamp)

class TimestampedObjectsList(list):
    """Handle timestamped object into a list.

    !!! warning "Timestamped objects are not sorted internally"
        
        Timestamped objects are considered to be stored according at their coming time.
    """

    def __init__(self, ts_object_type: type, ts_objects: list = []):

        self.__object_type = ts_object_type
        self.__object_properties = properties(self.__object_type)

        for ts_object in ts_objects:

            self.append(ts_object)

    @property
    def object_type(self):
        """Get object type handled by the list."""
        return self.__object_type

    def append(self, ts_object: TimestampedObjectType|dict):
        """Append timestamped object."""
        
        # Convert dict into GazePosition
        if type(ts_object) == dict:

            ts_object = self.__object_type.from_dict(ts_object)

        # Check object type
        if type(ts_object) != self.__object_type:

            if not issubclass(ts_object.__class__, self.__object_type):

                raise TypeError(f'{type(ts_object)} object is not {self.__object_type} instance')
        
        if not ts_object.is_timestamped():

            raise ValueError(f'object is not timestamped')
        
        super().append(ts_object)

    def look_for(self, timestamp: TimeStampType) -> TimestampedObjectType:
        """Look for object at given timestamp."""
        for ts_object in self:
            
            if ts_object.timestamp == timestamp:

                return ts_object

    def __add__(self, ts_objects: list = []) -> TimestampedObjectsListType:
        """Append timestamped objects list."""

        for ts_object in ts_objects:

            self.append(ts_object)

        return self

    @property
    def duration(self):
        """Get inferred duration from first and last timestamps."""
        if self:

            return self[-1].timestamp - self[0].timestamp

        else:

            return 0

    def timestamps(self):
        """Get all timestamps in list."""
        return [ts_object.timestamp for ts_object in self]

    def tuples(self) -> list:
        """Get all timestamped objects as list of tuple."""
        return [tuple(as_dict(ts_object, filter=False).values()) for ts_object in self]

    @classmethod
    def from_dataframe(self, ts_object_type: type, dataframe: pandas.DataFrame, exclude=[]) -> TimestampedObjectsListType:
        """Create a TimestampedObjectsList from [Pandas DataFrame](https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.html)."""

        dataframe.drop(exclude, inplace=True, axis=True)

        assert(dataframe.index.name == 'timestamp')

        object_list = [ts_object_type(timestamp=timestamp, **object_dict) for timestamp, object_dict in dataframe.to_dict('index').items()]

        return TimestampedObjectsList(ts_object_type, object_list)

    def as_dataframe(self, exclude=[], split={}) -> pandas.DataFrame:
        """Convert as [Pandas DataFrame](https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.html).  
        
        The optional *split* argument allows tuple values to be stored in dedicated columns.  
        For example: to convert {"point": (0, 0)} data as two separated "x" and "y" columns, use split={"point": ["x", "y"]}  

        !!! warning "Values must be dictionaries"
        
            Each key is stored as a column name.

        !!! note

            Timestamps are stored as index column called 'timestamp'.
        """

        df = pandas.DataFrame(self.tuples(), columns=self.__object_properties)

        # Exclude columns
        df.drop(exclude, inplace=True, axis=True)

        # Split columns
        if len(split) > 0:

            splited_columns = []
            
            for column in df.columns:

                if column in split.keys():

                    df[split[column]] = pandas.DataFrame(df[column].tolist(), index=df.index)
                    df.drop(column, inplace=True, axis=True)

                    for new_column in split[column]:

                        splited_columns.append(new_column)

                else:

                    splited_columns.append(column)

            # Reorder splited columns
            df = df[splited_columns]

        # Append timestamps as index column
        df['timestamp'] = self.timestamps()
        df.set_index('timestamp', inplace=True)

        return df

    @classmethod
    def from_json(self, ts_object_type: type, json_filepath: str) -> TimestampedObjectsListType:
        """Create a TimestampedObjectsList from .json file."""

        with open(json_filepath, encoding='utf-8') as ts_objects_file:

            json_ts_objects = json.load(ts_objects_file)

            return TimestampedObjectsList(ts_object_type, [ts_object_type(**ts_object_dict) for ts_object_dict in json_ts_objects])

    def to_json(self, json_filepath: str):
        """Save a TimestampedObjectsList to .json file."""

        with open(json_filepath, 'w', encoding='utf-8') as ts_objects_file:

            json.dump(self, ts_objects_file, ensure_ascii=False, default=(lambda obj: as_dict(obj)), indent='   ')

    def __repr__(self):
        """String representation"""
        return json.dumps([as_dict(ts_object) for ts_object in self], ensure_ascii=False,)

    def __str__(self):
        """String representation"""
        return json.dumps([as_dict(ts_object) for ts_object in self], ensure_ascii=False,)

    def pop_last_until(self, timestamp: TimeStampType) -> TimestampedObjectType:
        """Pop all item until a given timestamped value and return the first after."""

        # get last item before given timestamp
        earliest_value = self.get_last_until(timestamp)

        while self[0].timestamp < earliest_value.timestamp:

            self.pop(0)
            
        return self[0]

    def pop_last_before(self, timestamp: TimeStampType) -> TimestampedObjectType:
        """Pop all item before a given timestamped value and return the last one."""

        # get last item before given timestamp
        earliest_value = self.get_last_before(timestamp)

        poped_value = self.pop(0)

        while poped_value.timestamp != earliest_value.timestamp:

            poped_value = self.pop(0)

        return poped_value

    def get_first_from(self, timestamp: TimeStampType) -> TimestampedObjectType:
        """Retreive first item timestamp from a given timestamp value."""

        ts_list = self.timestamps()
        first_from_index = bisect.bisect_left(ts_list, timestamp)

        if first_from_index < len(self):

            return self[ts_list[first_from_index]]
            
        else:
            
            raise KeyError(f'No data stored after {timestamp} timestamp.')

    def get_last_before(self, timestamp: TimeStampType) -> TimestampedObjectType:
        """Retreive last item timestamp before a given timestamp value."""

        ts_list = self.timestamps()
        last_before_index = bisect.bisect_left(ts_list, timestamp) - 1

        if last_before_index >= 0:

            return self[ts_list[last_before_index]]
            
        else:
            
            raise KeyError(f'No data stored before {timestamp} timestamp.')
        
    def get_last_until(self, timestamp: TimeStampType) -> TimestampedObjectType:
        """Retreive last item timestamp until a given timestamp value."""

        ts_list = self.timestamps()
        last_until_index = bisect.bisect_right(ts_list, timestamp) - 1

        if last_until_index >= 0:

            return self[ts_list[last_until_index]]
            
        else:
            
            raise KeyError(f'No data stored until {timestamp} timestamp.')

    def plot(self, names=[], colors=[], split={}, samples=None) -> list:
        """Plot as [matplotlib](https://matplotlib.org/) time chart."""

        df = self.as_dataframe(split=split)
        legend_patches = []

        # decimate data
        if samples != None:

            if samples < len(df):

                step = int(len(df) / samples) + 1
                df = df.iloc[::step, :]

        for name, color in zip(names, colors):

            markerline, stemlines, baseline = mpyplot.stem(df.index, df[name])
            mpyplot.setp(markerline, color=color, linewidth=1, markersize = 1)
            mpyplot.setp(stemlines, color=color, linewidth=1)
            mpyplot.setp(baseline, color=color, linewidth=1)

            legend_patches.append(mpatches.Patch(color=color, label=name.upper()))

        return legend_patches

class SharedObject(TimestampedObject):
    """Abstract class to enable multiple threads sharing for timestamped object."""

    def __init__(self, timestamp: int|float = math.nan):

        TimestampedObject.__init__(self, timestamp)
        self._lock = threading.Lock()
        self._execution_times = {}
        self._exceptions = {}

class PipelineStepObject():
    """
    Define class to assess pipeline step methods execution time and observe them.
    """

    def __init__(self, name: str = None, working_directory: str = None, observers: dict = None):
        """Initialize PipelineStepObject

        Parameters:
            name: object name
            working_directory: folder path to use for relative file path.
            observers: dictionary with observers objects.

        """

        # Init private attribute
        self.__name = name
        self.__working_directory = working_directory
        self.__observers = observers if observers is not None else {}
        self.__execution_times = {}
        self.__properties = {}

        # parent attribute will be setup later by parent it self
        self.__parent = None

    def __enter__(self):
        """At with statement start."""

        # Start children pipeline step objects
        for child in self.children:

            child.__enter__()

        # Start observers
        for observer_name, observer in self.__observers.items():

            observer.__enter__()

        return self

    def __exit__(self, exception_type, exception_value, exception_traceback):
        """At with statement end."""

        # End observers
        for observer_name, observer in self.__observers.items():

            observer.__exit__(exception_type, exception_value, exception_traceback)

        # End children pipeline step objects
        for child in self.children:

            child.__exit__(exception_type, exception_value, exception_traceback)

    @property
    def name(self) -> str:
        """Get pipeline step object's name."""
        return self.__name

    @property
    def working_directory(self) -> str:
        """Get pipeline step object's working_directory."""
        return self.__working_directory

    @property
    def parent(self) -> object:
        """Get pipeline step object's parent object."""
        return self.__parent

    @parent.setter
    def parent(self, parent: object):
        """Set layer's parent object."""
        self.__parent = parent

    @property
    def observers(self) -> dict:
        """Get pipeline step object observers dictionary."""
        return self.__observers

    @property
    def execution_times(self):
        """Get pipeline step object observers execution times dictionary."""
        return self.__execution_times
    
    def as_dict(self) -> dict:
        """Export PipelineStepObject attributes as dictionary.

        Returns:
            object_data: dictionary with pipeline step object attributes values.
        """
        return {
            "name": self.__name,
            "observers": self.__observers
        }

    @classmethod
    def from_dict(cls, object_data: dict, working_directory: str = None) -> object:
        """Load PipelineStepObject attributes from dictionary.

        Returns:
            object_data: dictionary with pipeline step object attributes values.
            working_directory: folder path where to load files when a dictionary value is a relative filepath.
        """

        # Append working directory to the Python path
        if working_directory is not None:

            sys.path.append(working_directory)

        # Load name
        try:

            new_name = object_data.pop('name')

        except KeyError:

            new_name = None
        
        # Load observers
        new_observers = {}

        try:

            new_observers_value = object_data.pop('observers')

            # str: relative path to file
            if type(new_observers_value) == str:

                filepath = os.path.join(working_directory, new_observers_value)
                file_format = filepath.split('.')[-1]

                # Python file format
                if file_format == 'py':

                    observer_module_path = new_observers_value.split('.')[0]

                    observer_module = importlib.import_module(observer_module_path)

                    new_observers = observer_module.__observers__

             # dict: instanciate ready-made argaze observers
            elif type(new_observers_value) == dict:

                for observer_type, observer_data in new_observers_value.items():

                    new_observers[observer_type] = eval(f'{observer_type}(**observer_data)')

        except KeyError:

            pass

        # Create pipeline step object
        return PipelineStepObject(\
            new_name, \
            working_directory, \
            new_observers \
            )

    @classmethod
    def from_json(cls, configuration_filepath: str, patch_filepath: str = None) -> object:
        """
        Load pipeline step object from .json file.

        Parameters:
            configuration_filepath: path to json configuration file
            patch_filepath: path to json patch file to modify any configuration entries
        """
        with open(configuration_filepath) as configuration_file:

            object_data = json.load(configuration_file)
            working_directory = os.path.dirname(configuration_filepath)

            # Apply patch to configuration if required
            if patch_filepath is not None:

                with open(patch_filepath) as patch_file:

                    patch_data = json.load(patch_file)

                    import collections.abc

                    def update(d, u):

                        for k, v in u.items():

                            if isinstance(v, collections.abc.Mapping):

                                d[k] = update(d.get(k, {}), v)

                            elif v is None:

                                del d[k]

                            else:

                                d[k] = v

                        return d

                    object_data = update(object_data, patch_data)

            return cls.from_dict(object_data, working_directory)

    def to_json(self, json_filepath: str = None):
        """Save pipeline step object into .json file."""

        # Remember file path to ease rewriting
        if json_filepath is not None:

            self.__json_filepath = json_filepath

        # Open file
        with open(self.__json_filepath, 'w', encoding='utf-8') as object_file:

            json.dump({module_path(self):as_dict(self)}, object_file, ensure_ascii=False, indent=4)

            # QUESTION: maybe we need two saving mode?
            #json.dump(self, object_file, ensure_ascii=False, indent=4, cls=DataFeatures.JsonEncoder)

    def __str__(self) -> str:
        """
        String representation of pipeline step object.
        
        Returns:
            String representation
        """
        
        tabs = self.tabulation  
        output = f'{Fore.GREEN}{Style.BRIGHT}{self.__class__.__module__}.{self.__class__.__name__}{Style.RESET_ALL}\n'

        if self.__name is not None:
            output += f'{tabs}\t{Style.BRIGHT}name{Style.RESET_ALL}: {self.__name}\n'

        if self.__parent is not None:
            output += f'{tabs}\t{Style.BRIGHT}parent{Style.RESET_ALL}: {self.__parent.name}\n'

        if len(self.__observers):
            output += f'{tabs}\t{Style.BRIGHT}observers{Style.RESET_ALL}:\n'
            for name, observer in self.__observers.items():
                output += f'{tabs}\t  - {Fore.RED}{name}{Style.RESET_ALL}: {Fore.GREEN}{Style.BRIGHT}{observer.__class__.__module__}.{observer.__class__.__name__}{Style.RESET_ALL}\n'

        for name, value in self.properties:

            output += f'{tabs}\t{Style.BRIGHT}{name}{Style.RESET_ALL}: '

            if type(value) == dict:

                output += '\n'

                for k, v in value.items():

                    output += f'{tabs}\t  - {Fore.RED}{k}{Style.RESET_ALL}: {v}\n'

            elif type(value) == numpy.ndarray:

                output += f'numpy.array{value.shape}\n'

            elif type(value) == pandas.DataFrame:

                output += f'pandas.DataFrame{value.shape}\n'

            else:

                output += f'{value}'

                if output[-1] != '\n':

                    output += '\n'

        return output

    @property
    def tabulation(self) -> str:
        """Edit tabulation string according parents number."""

        tabs = ''
        parent = self.__parent

        while (parent is not None):

            tabs += '\t'
            parent = parent.parent

        return tabs

    @property
    def properties(self) -> Tuple[name, any]:
        """Iterate over pipeline step properties values."""

        for name, item in self.__class__.__dict__.items(): 

            if isinstance(item, property):

                yield name, getattr(self, name)
        
        for base in self.__class__.__bases__:

            if base != PipelineStepObject and base != SharedObject:

                for name, item in base.__dict__.items(): 

                    if isinstance(item, property):

                        yield name, getattr(self, name)

    @property
    def children(self) -> object:
        """Iterate over children pipeline step objects."""

        for name, item in self.__class__.__dict__.items(): 

            if isinstance(item, property):

                attr = getattr(self, name)

                if PipelineStepObject in attr.__class__.__bases__:

                    yield attr

def PipelineStepMethod(method):
    """Define a decorator use into PipelineStepObject class to declare pipeline method.

    !!! danger
        PipelineStepMethod must have a timestamp as first argument.
    """

    def wrapper(self, *args, timestamp: int|float = None, unwrap: bool = False, **kwargs):
        """Wrap pipeline step method to measure execution time.

        Parameters:
            args: Any arguments defined by PipelineStepMethod.
            timestamp: Optional method call timestamp (unit does'nt matter) if first args parameter is not a TimestampedObject instance.
            unwrap: Extra arguments used in wrapper function to call wrapped method directly.
        """
        if timestamp is None:

            if isinstance(args[0], TimestampedObject):

                timestamp = args[0].timestamp

        if unwrap:

            return method(self, *args, **kwargs)

        # Initialize execution time assessment
        start = time.perf_counter()
        exception = None
        result = None
        
        try:

            # Execute wrapped method
            result = method(self, *args, **kwargs)

        except Exception as e:

            exception = e

        finally:

            # Measure execution time
            self.execution_times[method.__name__] = (time.perf_counter() - start) * 1e3

        # Notify observers that method has been called
        subscription_name = f'on_{method.__name__}'

        for observer_name, observer in self.observers.items():

            # Does the observer cares about this method?
            if subscription_name in dir(observer): 

                subscription = getattr(observer, subscription_name)

                # Call subscription
                subscription(timestamp, self, exception)

        # Raise exception
        if exception is not None:

            raise exception

        return result

    return wrapper

class PipelineStepObserver():
    """Define abstract class to observe pipeline step object use.

    !!! note
        To subscribe to a method call, the inherited class simply needs to define 'on_<method_name>' functions with timestamp, object and traceback argument.
    """

    def __enter__(self):
        """
        Define abstract __enter__ method to use observer as a context.

        !!! warning
            This method is called provided that the observed PipelineStepObject is created as a context using a with statement.
        """
        return self

    def __exit__(self, type, value, traceback):
        """
        Define abstract __exit__ method to use observer as a context.

        !!! warning
            This method is called provided that the observed PipelineStepObject is created as a context using a with statement.
        """
        pass