generate_data.py

# ref code
# https://github.com/guacomolia/ptr_net/blob/master/generate_data.py

import numpy as np
import random


def generate_single_seq(length=30, min_len=5, max_len=10):
    # https://medium.com/@devnag/pointer-networks-in-tensorflow-with-sample-code-14645063f264
    """ Generates a sequence of numbers of random length and inserts a sub-sequence oh greater numbers at random place
    Input:
    length: total sequence length
    min_len: minimum length of sequence
    max_len: maximum length of sequence
    Output: Sequence of numbers, index of the start of greater numbers subsequence"""
    seq_before = [(random.randint(1, 5)) for _ in range(random.randint(min_len, max_len))]
    seq_during = [(random.randint(6, 10)) for _ in range(random.randint(min_len, max_len))]
    seq_after = [random.randint(1, 5) for _ in range(random.randint(min_len, max_len))]
    seq = seq_before + seq_during + seq_after
    seq = seq + ([0] * (length - len(seq)))
    return (seq, len(seq_before), len(seq_before) + len(seq_during)-1)


def generate_set_seq(N):
    """
    The `generate_set_seq` function generates a set of N sequences of fixed length for `Boundary tasks`.
    It returns the data, starts and ends lists.
    The data list contains all the sequences in string format.
    The starts list contains all the starting indices for each sequence in integer format, and similarly for ends.

    :param N: Generate n sequences
    :return: A list of sequences, a list of starting indices and a list of ending indices
    """
    data = []
    starts = []
    ends = []
    for _ in range(N):
        seq, ind_start, ind_end = generate_single_seq()
        data.append(seq)
        starts.append(ind_start)
        ends.append(ind_end)
    return data, starts, ends


def make_seq_data(n_samples, seq_len):
    # Boundary tasks
    data, labels = [], []
    for _ in range(n_samples):
        input = np.random.permutation(range(seq_len)).tolist()
        target = sorted(range(len(input)), key=lambda k: input[k])
        data.append(input)
        labels.append(target)
    return data, labels