molnet file

"""

BBBC Dataset loader.

This file contains image loaders for the BBBC dataset collection (https://data.broadinstitute.org/bbbc/image_sets.html).

"""

from __future__ import division

from __future__ import unicode_literals

import os

import numpy as np

import logging

import deepchem

logger = logging.getLogger(__name__)

def load_bbbc001(split='index', reload=True):

  """Load BBBC001 dataset

  This dataset contains 6 images of human HT29 colon cancer cells. The task is to learn to predict the cell counts in these images. This dataset is too small to serve to train algorithms, but might serve as a good test dataset. https://data.broadinstitute.org/bbbc/BBBC001/

  """

  # Featurize BBBC001 dataset

  bbbc001_tasks = ["cell-count"]

  data_dir = deepchem.utils.get_data_dir()

  if reload:

    save_dir = os.path.join(data_dir, "bbbc001/" + str(split))

    loaded, all_dataset, transformers = deepchem.utils.save.load_dataset_from_disk(

        save_dir)

    if loaded:

      return bbbc001_tasks, all_dataset, transformers

  dataset_file = os.path.join(data_dir, "BBBC001_v1_images_tif.zip")

  labels_file = os.path.join(data_dir, "BBBC001_v1_counts.txt")

  if not os.path.exists(dataset_file):

    deepchem.utils.download_url(

        'https://data.broadinstitute.org/bbbc/BBBC001/BBBC001_v1_images_tif.zip'

    )

  if not os.path.exists(labels_file):

    deepchem.utils.download_url(

        'https://data.broadinstitute.org/bbbc/BBBC001/BBBC001_v1_counts.txt'

    )

  # Featurize Images into NumpyArrays

  loader = deepchem.data.ImageLoader()

  dataset = loader.featurize(dataset_file, in_memory=False)

  # Load text file with labels

  with open(labels_file) as f:

    content = f.readlines()

  # Strip the first line which holds field labels

  lines = [x.strip() for x in content][1:]

  # Format is: Image_name count1 count2 

  lines = [x.split("\t") for x in lines]

  counts = [(float(x[1]) + float(x[2]))/2.0 for x in lines]

  y = np.array(counts)

  # This is kludgy way to add y to dataset. Can be done better?

  dataset = deepchem.data.DiskDataset.from_numpy(dataset.X, y)

  if split == None:

    return bbbc001_tasks, (dataset, None, None), transformers

  splitters = {

      'index': deepchem.splits.IndexSplitter(),

      'random': deepchem.splits.RandomSplitter(),

  }

  if split not in splitters:

    raise ValueError("Only index and random splits supported.")

  splitter = splitters[split]

  train, valid, test = splitter.train_valid_test_split(dataset)

  all_dataset = (train, valid, test)

  if reload:

    deepchem.utils.save.save_dataset_to_disk(save_dir, train, valid, test,

                                             transformers)

  return bbbc001_tasks, all_dataset, transformers