# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""TODO: Add a description here."""

import statistics
from typing import Optional, Union

import datasets
import evaluate
import numpy as np

# TODO: Add BibTeX citation
_CITATION = """\
@InProceedings{huggingface:module,
title = {A great new module},
authors={huggingface, Inc.},
year={2020}
}
"""

# TODO: Add description of the module here
_DESCRIPTION = """\
This new module is designed to solve this great ML task and is crafted with a lot of care.
"""


# TODO: Add description of the arguments of the module here
_KWARGS_DESCRIPTION = """
Calculates how good are predictions given some references, using certain scores
Args:
    predictions: list of generated time series.
        shape: (num_generation, num_timesteps, num_features)
    references: list of reference
        shape: (num_reference, num_timesteps, num_features)
Returns:
Examples:
    Examples should be written in doctest format, and should illustrate how
    to use the function.

    >>> my_new_module = evaluate.load("bowdbeg/matching_series")
    >>> results = my_new_module.compute(references=[[[0.0, 1.0]]], predictions=[[[0.0, 1.0]]])
    >>> print(results)
    {'matchin': 1.0}
"""


@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
class matching_series(evaluate.Metric):
    """TODO: Short description of my evaluation module."""

    def _info(self):
        # TODO: Specifies the evaluate.EvaluationModuleInfo object
        return evaluate.MetricInfo(
            # This is the description that will appear on the modules page.
            module_type="metric",
            description=_DESCRIPTION,
            citation=_CITATION,
            inputs_description=_KWARGS_DESCRIPTION,
            # This defines the format of each prediction and reference
            features=datasets.Features(
                {
                    "predictions": datasets.Sequence(datasets.Sequence(datasets.Value("float"))),
                    "references": datasets.Sequence(datasets.Sequence(datasets.Value("float"))),
                }
            ),
            # Homepage of the module for documentation
            homepage="https://huggingface.co/spaces/bowdbeg/matching_series",
            # Additional links to the codebase or references
            codebase_urls=["http://github.com/path/to/codebase/of/new_module"],
            reference_urls=["http://path.to.reference.url/new_module"],
        )

    def _download_and_prepare(self, dl_manager):
        """Optional: download external resources useful to compute the scores"""
        pass

    def compute(self, *, predictions=None, references=None, **kwargs) -> Optional[dict]:
        """Compute the evaluation module.

        Usage of positional arguments is not allowed to prevent mistakes.

        Args:
            predictions (`list/array/tensor`, *optional*):
                Predictions.
            references (`list/array/tensor`, *optional*):
                References.
            **kwargs (optional):
                Keyword arguments that will be forwarded to the evaluation module [`~evaluate.EvaluationModule.compute`]
                method (see details in the docstring).

        Return:
            `dict` or `None`

            - Dictionary with the results if this evaluation module is run on the main process (`process_id == 0`).
            - `None` if the evaluation module is not run on the main process (`process_id != 0`).

        ```py
        >>> import evaluate
        >>> accuracy =  evaluate.load("accuracy")
        >>> accuracy.compute(predictions=[0, 1, 1, 0], references=[0, 1, 0, 1])
        ```
        """
        all_kwargs = {"predictions": predictions, "references": references, **kwargs}
        if predictions is None and references is None:
            missing_kwargs = {k: None for k in self._feature_names() if k not in all_kwargs}
            all_kwargs.update(missing_kwargs)
        else:
            missing_inputs = [k for k in self._feature_names() if k not in all_kwargs]
            if missing_inputs:
                raise ValueError(
                    f"Evaluation module inputs are missing: {missing_inputs}. All required inputs are {list(self._feature_names())}"
                )
        inputs = {input_name: all_kwargs[input_name] for input_name in self._feature_names()}
        compute_kwargs = {k: kwargs[k] for k in kwargs if k not in self._feature_names()}
        return self._compute(**inputs, **compute_kwargs)

    def _compute(
        self,
        predictions: Union[list, np.ndarray],
        references: Union[list, np.ndarray],
        batch_size: Optional[int] = None,
    ):
        """
        Compute the scores of the module given the predictions and references
        Args:
            predictions: list of generated time series.
                shape: (num_generation, num_timesteps, num_features)
            references: list of reference
                shape: (num_reference, num_timesteps, num_features)
            batch_size: batch size to use for the computation. If None, the whole dataset is processed at once.
        Returns:
        """
        predictions = np.array(predictions)
        references = np.array(references)
        if predictions.shape[1:] != references.shape[1:]:
            raise ValueError(
                "The number of features in the predictions and references should be the same. predictions: {}, references: {}".format(
                    predictions.shape[1:], references.shape[1:]
                )
            )

        # at first, convert the inputs to numpy arrays

        # MSE between predictions and references for all example combinations for each features
        # shape: (num_generation, num_reference, num_features)
        if batch_size is not None:
            mse = np.zeros((len(predictions), len(references), predictions.shape[-1]))
            # iterate over the predictions and references in batches
            for i in range(0, len(predictions) + batch_size, batch_size):
                for j in range(0, len(references) + batch_size, batch_size):
                    mse[i : i + batch_size, j : j + batch_size] = np.mean(
                        (predictions[i : i + batch_size, None] - references[None, j : j + batch_size]) ** 2, axis=-2
                    )
        else:
            mse = np.mean((predictions[:, None] - references) ** 2, axis=1)

        index_mse = mse.diagonal(axis1=0, axis2=1).mean()

        # matching scores
        mse_mean = mse.mean(axis=-1)
        # best match for each generated time series
        # shape: (num_generation,)
        best_match = np.argmin(mse_mean, axis=-1)

        # matching mse
        # shape: (num_generation,)
        matching_mse = mse_mean[np.arange(len(best_match)), best_match].mean()

        # best match for each reference time series
        # shape: (num_reference,)
        best_match_inv = np.argmin(mse_mean, axis=0)
        covered_mse = mse_mean[best_match_inv, np.arange(len(best_match_inv))].mean()

        harmonic_mean = 2 / (1 / matching_mse + 1 / covered_mse)

        # take matching for each feature and compute metrics for them
        matching_mse_features = []
        covered_mse_features = []
        harmonic_mean_features = []
        index_mse_features = []
        for f in range(predictions.shape[-1]):
            mse_f = mse[:, :, f]
            index_mse_f = mse_f.diagonal(axis1=0, axis2=1).mean()
            best_match_f = np.argmin(mse_f, axis=-1)
            matching_mse_f = mse_f[np.arange(len(best_match_f)), best_match_f].mean()
            best_match_inv_f = np.argmin(mse_f, axis=0)
            covered_mse_f = mse_f[best_match_inv_f, np.arange(len(best_match_inv_f))].mean()
            harmonic_mean_f = 2 / (1 / matching_mse_f + 1 / covered_mse_f)
            matching_mse_features.append(matching_mse_f)
            covered_mse_features.append(covered_mse_f)
            harmonic_mean_features.append(harmonic_mean_f)
            index_mse_features.append(index_mse_f)
        macro_matching_mse = statistics.mean(matching_mse_features)
        macro_covered_mse = statistics.mean(covered_mse_features)
        macro_harmonic_mean = statistics.mean(harmonic_mean_features)
        macro_index_mse = statistics.mean(index_mse_features)
        return {
            "matching_mse": matching_mse,
            "harmonic_mean": harmonic_mean,
            "covered_mse": covered_mse,
            "index_mse": index_mse,
            "matching_mse_features": matching_mse_features,
            "harmonic_mean_features": harmonic_mean_features,
            "covered_mse_features": covered_mse_features,
            "index_mse_features": index_mse_features,
            "macro_matching_mse": macro_matching_mse,
            "macro_covered_mse": macro_covered_mse,
            "macro_harmonic_mean": macro_harmonic_mean,
            "macro_index_mse": macro_index_mse,
        }