# 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.
"""SBERT consime similarity metric."""

import evaluate
import datasets
import torch
import torch.nn as nn
from transformers import AutoTokenizer, BertModel

_CITATION = """\
@article{Reimers2019,
archivePrefix = {arXiv},
arxivId = {1908.10084},
author = {Reimers, Nils and Gurevych, Iryna},
doi = {10.18653/v1/d19-1410},
eprint = {1908.10084},
isbn = {9781950737901},
journal = {EMNLP-IJCNLP 2019 - 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint Conference on Natural Language Processing, Proceedings of the Conference},
pages = {3982--3992},
title = {{Sentence-BERT: Sentence embeddings using siamese BERT-networks}},
year = {2019}
}
"""

_DESCRIPTION = """\
Use SBERT to produce embedding and score the similarity by cosine similarity
"""


_KWARGS_DESCRIPTION = """
Calculates how semantic similarity are predictions given some references, using certain scores
Args:
    predictions: list of predictions to score. Each predictions
        should be a string with tokens separated by spaces.
    references: list of reference for each prediction. Each
        reference should be a string with tokens separated by spaces.
Returns:
    score: description of the first score,
Examples:
    Examples should be written in doctest format, and should illustrate how
    to use the function.

    >>> sbert_cosine = evaluate.load("transZ/sbert_cosine")
    >>> results = my_new_module.compute(references=["Nice to meet you"], predictions=["It is my pleasure to meet you"])
    >>> print(results)
    {'score': 0.85}
"""

@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
class sbert_cosine(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.Value("string", id="sequence"),
                        "references": datasets.Sequence(datasets.Value("string", id="sequence"), id="references"),
                    }
                ),
                datasets.Features(
                    {
                        "predictions": datasets.Value("string", id="sequence"),
                        "references": datasets.Value("string", id="sequence"),
                    }
                ),
            ],
            # Homepage of the module for documentation
            homepage="http://sbert.net",
            # Additional links to the codebase or references
            codebase_urls=["https://github.com/UKPLab/sentence-transformers"],
            reference_urls=["https://github.com/UKPLab/sentence-transformers"]
        )

    def _download_and_prepare(self, dl_manager):
        """Optional: download external resources useful to compute the scores"""
        # TODO: Download external resources if needed
        pass

    def _compute(self, predictions, references, model_type='sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2'):
        """Returns the scores"""
        def mean_pooling(model_output, attention_mask):
            token_embeddings = model_output[0] #First element of model_output contains all token embeddings
            input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
            return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
        
        def batch_to_device(batch, target_device):
            """
            send a pytorch batch to a device (CPU/GPU)
            """
            for key in batch:
                if isinstance(batch[key], torch.Tensor):
                    batch[key] = batch[key].to(target_device)
            return batch

        device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

        tokenizer = AutoTokenizer.from_pretrained(model_type)
        model = BertModel.from_pretrained(model_type)
        model = model.to(device)
        cosine = nn.CosineSimilarity(dim=0)

        def calculate(x: str, y: str):
            encoded_input = tokenizer([x, y], padding=True, truncation=True, return_tensors='pt')
            encoded_input = batch_to_device(encoded_input, device)
            model_output = model(**encoded_input)
            embeds = mean_pooling(model_output, encoded_input['attention_mask'])
            res = cosine(embeds[0, :], embeds[1, :]).item()
            return res

        # avg = lambda x: sum(x) / len(x)

        with torch.no_grad():
            scores = [calculate(pred, ref) for pred, ref in zip(predictions, references)]

        return {
            "score": scores,
        }