Use of Custom Readers that support Pandas DataFrames

This is a demonstration of using customised readers that support data contained within Pandas DataFrames, rather than loading directly from a .csv file using CSVGroundTruthReader or CSVDetectionReader.

The benefit is that this allows us to use the versatile data loading capabilities of pandas to read from many different data source types as needed, including .csv, JSON, XML, Parquet, HDF5, .txt, .zip and more. The resulting DataFrame can then simply be fed into the defined DataFrameGroundTruthReader or DataFrameDetectionReader for further processing in Stone Soup as required.

Software dependencies

Before beginning this example, you need to ensure that Pandas is installed, which is a fast, powerful and flexible open-source data analysis tool in Python. The easiest way to install pandas (if not done so already), is to run pip install from a terminal window within the desired environment:

pip install pandas

The main dependencies and imports for this example are included below:

import numpy as np
import pandas as pd

from math import modf

from stonesoup.base import Property
from stonesoup.buffered_generator import BufferedGenerator
from stonesoup.reader.base import GroundTruthReader, DetectionReader, Reader
from stonesoup.types.detection import Detection
from stonesoup.types.groundtruth import GroundTruthPath, GroundTruthState

from typing import Sequence, Collection

from datetime import datetime, timedelta
from dateutil.parser import parse

Data Frame Reader

Similarly to Stone Soup’s _CSVFrameReader, we’ll define a _DataFrameReader class that inherits from the base Reader class to read a DataFrame containing state vector fields, a time field, and additional metadata fields (all other columns by default). The only difference between this class and the _CSVFrameReader class is that we have no path attribute (the DataFrame is already loaded in memory).

class _DataFrameReader(Reader):
    state_vector_fields: Sequence[str] = Property(
        doc='List of columns names to be used in state vector')
    time_field: str = Property(
        doc='Name of column to be used as time field')
    time_field_format: str = Property(
        default=None, doc='Optional datetime format')
    timestamp: bool = Property(
        default=False, doc='Treat time field as a timestamp from epoch')
    metadata_fields: Collection[str] = Property(
        default=None, doc='List of columns to be saved as metadata, default all')

    def _get_metadata(self, row):
        if self.metadata_fields is None:
            local_metadata = dict(row)
            for key in list(local_metadata):
                if key == self.time_field or key in self.state_vector_fields:
                    del local_metadata[key]
        else:
            local_metadata = {field: row[field]
                              for field in self.metadata_fields
                              if field in row}
        return local_metadata

    def _get_time(self, row):
        if self.time_field_format is not None:
            time_field_value = datetime.strptime(row[self.time_field], self.time_field_format)
        elif self.timestamp:
            fractional, timestamp = modf(float(row[self.time_field]))
            time_field_value = datetime.utcfromtimestamp(int(timestamp))
            time_field_value += timedelta(microseconds=fractional * 1E6)
        else:
            time_field_value = row[self.time_field]

            if not isinstance(time_field_value, datetime):
                time_field_value = parse(time_field_value, ignoretz=True)

        return time_field_value

Data Ground Truth Reader

With the help of our _DataFrameReader class, we can now define a custom DataFrameGroundTruthReader. This is similar to CSVGroundTruthReader and inherits from the base GroundTruthReader class. A key difference is that we include an instance attribute for the dataframe containing our data.

We also define a custom generator function (groundtruth_paths_gen) that uses the decorator @BufferedGenerator.generator_method. The generator needs to return a time and a set of detections, like so:

class DataFrameGroundTruthReader(GroundTruthReader, _DataFrameReader):
    """A custom reader for pandas DataFrames containing truth data.

    The DataFrame must have colums containing all fields needed to generate the
    ground truth state. Those states with the same ID will be put into
    a :class:`~.GroundTruthPath` in sequence, and all paths that are updated at the same time
    are yielded together, and such assumes file is in time order.

    Parameters
    ----------
    """
    dataframe: pd.DataFrame = Property(doc="DataFrame containing the ground truth data.")
    path_id_field: str = Property(doc='Name of column to be used as path ID')

    @BufferedGenerator.generator_method
    def groundtruth_paths_gen(self):
        """ Generator method for providing each row of ground truth data. """
        groundtruth_dict = {}
        updated_paths = set()
        previous_time = None
        for row in self.dataframe.to_dict(orient="records"):

            time = self._get_time(row)
            if previous_time is not None and previous_time != time:
                yield previous_time, updated_paths
                updated_paths = set()
            previous_time = time

            state = GroundTruthState(np.array([[row[col_name]] for col_name
                                              in self.state_vector_fields],
                                              dtype=np.float64),
                                     timestamp=time,
                                     metadata=self._get_metadata(row))

            id_ = row[self.path_id_field]
            if id_ not in groundtruth_dict:
                groundtruth_dict[id_] = GroundTruthPath(id=id_)
            groundtruth_path = groundtruth_dict[id_]
            groundtruth_path.append(state)
            updated_paths.add(groundtruth_path)

            # Yield remaining
        yield previous_time, updated_paths

With our DataFrameGroundTruthReader defined, we can test it on a simple example. Let’s do a basic 3D simulation to create an example dataframe, from which we can test our class:

from stonesoup.models.transition.linear import CombinedLinearGaussianTransitionModel, \
                                               ConstantVelocity

q_x = 0.05
q_y = 0.05
q_z = 0.05
start_time = datetime.now()
transition_model = CombinedLinearGaussianTransitionModel([ConstantVelocity(q_x),
                                                          ConstantVelocity(q_y),
                                                          ConstantVelocity(q_z)])
truth = GroundTruthPath([GroundTruthState([0, 1, 0, 1, 0, 1], timestamp=start_time)])

times = []
x, y, z = [], [], []
vel_x, vel_y, vel_z = [], [], []

num_steps = 25
for k in range(1, num_steps + 1):

    time = start_time+timedelta(seconds=k)

    next_state = GroundTruthState(
        transition_model.function(truth[k-1], noise=True,
                                  time_interval=timedelta(seconds=1)),
        timestamp=time)
    truth.append(next_state)

    times.append(time)
    x.append(next_state.state_vector[0])
    vel_x.append(next_state.state_vector[1])
    y.append(next_state.state_vector[2])
    vel_y.append(next_state.state_vector[3])
    z.append(next_state.state_vector[4])
    vel_z.append(next_state.state_vector[5])

truth_df = pd.DataFrame({'time': times,
                         'x': x,
                         'y': y,
                         'z': z,
                         'vel_x': vel_x,
                         'vel_y': vel_y,
                         'vel_z': vel_z,
                         'track_id': 0})

truth_df.head(5)
time x y z vel_x vel_y vel_z track_id
0 2024-06-14 10:00:29.782013 0.865338 0.957984 1.099910 0.748356 0.958708 1.303132 0
1 2024-06-14 10:00:30.782013 1.741403 1.921418 2.505441 0.961461 0.874579 1.492441 0
2 2024-06-14 10:00:31.782013 2.527232 2.714486 3.817204 0.627899 0.816046 1.246632 0
3 2024-06-14 10:00:32.782013 2.926628 3.614685 5.040452 0.202304 1.027160 1.238388 0
4 2024-06-14 10:00:33.782013 3.178518 4.499810 6.049363 0.130322 0.775230 0.888243 0


Note that the process above is just an example for providing a simple dataframe to use, and is not generally something we would need to do (since we already have the GroundTruthPath). The dataframe above is just used to show the workings of our newly defined DataFrameGroundTruthReader. In practice, we can use any dataframe containing our Cartesian positions or longitude and latitude co-ordinates. Note that if we are using longitude and latitude inputs, we would also need to transform these using LongLatToUTMConverter or equivalent.

We can now initialise our DataFrameGroundTruthReader using this example DataFrame like so:

# read ground truth data from pandas dataframe
ground_truth_reader = DataFrameGroundTruthReader(
    dataframe=truth_df,
    state_vector_fields=['x', 'vel_x', 'y', 'vel_y', 'z', 'vel_z'],
    time_field='time',
    path_id_field='track_id')

Let’s demonstrate the ground truth reader generating output for one iteration:

next(iter(ground_truth_reader))

Another benefit of this ground truth reader is that we now have convenient access to the original dataframe, using the .dataframe attribute, like so:

ground_truth_reader.dataframe.head(3)
time x y z vel_x vel_y vel_z track_id
0 2024-06-14 10:00:29.782013 0.865338 0.957984 1.099910 0.748356 0.958708 1.303132 0
1 2024-06-14 10:00:30.782013 1.741403 1.921418 2.505441 0.961461 0.874579 1.492441 0
2 2024-06-14 10:00:31.782013 2.527232 2.714486 3.817204 0.627899 0.816046 1.246632 0


DataFrame Detection Reader

Similarly to our DataFrameGroundTruthReader, we can develop a custom DataFrameDetectionReader that can read in DataFrames containing detections through subclassing from Stone Soup’s DetectionReader class, along with our custom _DataFrameReader class above. Again, this closely resembles the existing CSVDetectionReader class within the Stone Soup library, except we include a instance attribute ‘dataframe’, and modify our detections_gen function to work with dataframes rather than .csv files. This can be seen below:

class DataFrameDetectionReader(DetectionReader, _DataFrameReader):
    """A custom detection reader for DataFrames containing detections.

    DataFrame must have headers with the appropriate fields needed to generate
    the detection. Detections at the same time are yielded together, and such assume file is in
    time order.

    Parameters
    ----------
    """
    dataframe: pd.DataFrame = Property(doc="DataFrame containing the detection data.")

    @BufferedGenerator.generator_method
    def detections_gen(self):
        detections = set()
        previous_time = None
        for row in self.dataframe.to_dict(orient="records"):

            time = self._get_time(row)
            if previous_time is not None and previous_time != time:
                yield previous_time, detections
                detections = set()
            previous_time = time

            detections.add(Detection(
                np.array([[row[col_name]] for col_name in self.state_vector_fields],
                         dtype=np.float64),
                timestamp=time,
                metadata=self._get_metadata(row)))

        # Yield remaining
        yield previous_time, detections

We can instantiate this using our example DataFrame above like so:

detection_reader = DataFrameDetectionReader(
    dataframe=truth_df,
    state_vector_fields=['x', 'vel_x', 'y', 'vel_y', 'z', 'vel_z'],
    time_field='time')

Following this, we can now perform any desired follow-up task such as simulation or tracking as covered in the other Stone Soup examples, tutorials and demonstrations. As discussed previously, the huge benefits of using a custom DataFrame reader like this is that we can read any type of data supported by the pandas library, which gives us a huge range of options. This strategy also saves us the overhead of manually specifying custom Stone Soup Reader classes for each format of data.

Total running time of the script: (0 minutes 0.236 seconds)

Gallery generated by Sphinx-Gallery