#!/usr/bin/env python3
import os
import time
import click
import datetime
from tqdm import tqdm
import pandas as pd
import numpy as np
from biom import load_table, Table
from biom.util import biom_open
from skbio.stats.composition import clr, centralize, closure
from skbio.stats.composition import clr_inv as softmax
from scipy.stats import entropy, spearmanr
from scipy.sparse import csr_matrix
from mmvec.mmvec import run_mmvec
from mmvec.dataset import PairedDataset, split_tables
from torch.utils.data import DataLoader
from decimal import Decimal


@click.group()
def mmvec():
    pass


@mmvec.command()
@click.option('--input-abundances',
              help=('Input abundances to use for prediction. '
                    'These could be metabolite intensities or sequence counts.'
              ))
@click.option('--output-abundances',
              help=('Output abundances to predict from input abundances. '
                    'These could be metabolite intensities or sequence counts.'
              ))
@click.option('--metadata-file', default=None,
              help='Input sample metadata file')
@click.option('--training-column',
              help=('Column in the sample metadata specifying which '
                    'samples are for training and testing.'),
              default=None)
@click.option('--num-testing-examples',
              help=('Number of samples to randomly select for testing'),
              default=10)
@click.option('--min-feature-count',
              help=('Minimum number of samples a microbe needs to be observed '
                    'in order to not filter out'),
              default=10)
@click.option('--epochs',
              help=('Number of epochs to train, one epoch represents the '
                    'number of samples to process an entire dataset.'), default=10)
@click.option('--batch-size',
              help='Number of samples to analyze per iteration.', default=10)
@click.option('--latent-dim',
              help=('Dimensionality of shared latent space. '
                    'This is analogous to the number of PC axes.'),
              default=3)
@click.option('--input-prior',
              help=('Scale of the normal prior on the input factor matrix. '
                    'Smaller prior will yield stronger regularization.'),
              default=1)
@click.option('--output-prior',
              help=('Scale of the normal prior on the output factor matrix. '
                    'Smaller prior will yield stronger regularization.'),
              default=1)
@click.option('--arm-the-gpu', is_flag=True,
              help=('Enables GPU support'),
              default=False)
@click.option('--num-workers',
              help=('Number of processes for preprocessing tables.'),
              default=1)
@click.option('--learning-rate',
              help=('Gradient descent learning rate.'),
              default=1e-1)
@click.option('--num-steps',
              help=('Number of steps for learning rate decay.'),
              default=10)
@click.option('--decay-rate',
              help=('Decay rate for learning'),
              default=0.1)
@click.option('--beta1',
              help=('Gradient decay rate for first Adam momentum estimates'),
              default=0.9)
@click.option('--beta2',
              help=('Gradient decay rate for second Adam momentum estimates'),
              default=0.95)
@click.option('--clip-norm',
              help=('Normalization constant for gradient clipping.'),
              default=10)
@click.option('--seed',
              help=('Random seed for the sake of reproducibility (optional).'),
              default=None)
@click.option('--checkpoint-interval',
              help=('Number of seconds before a storing a checkpoint.'),
              default=1000)
@click.option('--summary-dir', default='summarydir',
              help='Summary directory to save cross validation results.')
@click.option('--embeddings-file', default=None,
              help=('Path to save the embeddings learned from the model. '
                    'If this is not specified, then this will be saved under '
                    '`--summary-dir`.'))
@click.option('--ranks-file', default=None,
              help=('Path to save the ranks learned from the model. '
                    'If this is not specified, then this will be saved under '
                    '`--summary-dir`.'))
@click.option('--ordination-file', default=None,
              help=('Path to save the ordination learned from the model. '
                    'If this is not specified, then this will be saved under '
                    '`--summary-dir`.'))
def mmvec(input_abundances, output_abundances,
          metadata_file, training_column,
          num_testing_examples, min_feature_count,
          epochs, batch_size,  latent_dim,
          input_prior, output_prior,
          arm_the_gpu, num_workers, learning_rate,
          num_steps, decay_rate,
          beta1, beta2, clip_norm, seed,
          checkpoint_interval, summary_dir,
          embeddings_file, ranks_file, ordination_file):
    # set random seed if specified
    if seed is not None:
        np.random.seed(seed)
        torch.manual_seed(seed)

    microbes = load_table(input_abundances)
    metabolites = load_table(output_abundances)

    if metadata_file is not None:
        metadata = pd.read_table(metadata_file, index_col=0)
    else:
        metadata = None

    embeds, ranks, ordination = run_mmvec(
        microbes=microbes, metabolites=metabolites,
        metadata=metadata,
        training_column=training_column,
        num_testing_examples=num_testing_examples,
        min_feature_count=min_feature_count,
        epochs=epochs,
        batch_size=batch_size,
        latent_dim=latent_dim,
        input_prior=input_prior,
        output_prior=output_prior,
        beta1=0.9, beta2=0.99,
        num_workers=num_workers,
        clip_norm=clip_norm,
        num_steps=num_steps,
        decay_rate=decay_rate,
        learning_rate=learning_rate,
        arm_the_gpu=arm_the_gpu,
        checkpoint_interval=checkpoint_interval,
        summary_dir=summary_dir)

    files = []
    for f, ext in [(embeddings_file, 'embeddings.txt'),
                   (ranks_file, 'ranks.csv'),
                   (ordination_file, 'ordination.txt')]:
        if f is None:
            sname = '_'.join(
                ['log_PC%d' % latent_dim,
                 'lr%s' % "{:.2E}".format(Decimal(learning_rate)),
                 'ipr%.2f' % input_prior,
                 'opr%.2f' % output_prior,
                 'b%.2f' % beta1,
                 'bb%.2f' % beta2]
            )
            files.append(os.path.join(summary_dir, sname + '_' + ext))

    embeds.to_csv(files[0])
    ranks.to_csv(files[1])
    ordination.write(files[2])


if __name__ == '__main__':
    mmvec()
