philip-bl

philip-bl

# becomes red if previous command's result is not zero
# records time
# looks kinda nice

function red_if_nonzero {
  RETVAL=$?;  # get status code of the previously run command
  [ $RETVAL -ne 0 ] && tput setaf 1;  # if it's not 0, use color red
  return 0;
}

philip-bl

philip-bl

#include <cstddef>
#include <cstdint>
#include <list>
#include <vector>
#include <functional> // use std::hash<Key>
#include <memory> // unique_ptr
#include <stdexcept> // out_of_range
#include <iostream>
#include <string>
#include <unordered_map>

philip-bl

# https://about.sourcegraph.com/blog/thyme-a-simple-cli-to-measure-human-time-and-focus/
# https://github.com/sourcegraph/thyme

[Unit]
Description=Call thyme to save windows usage to json for statistics

[Service]
Type=oneshot
ExecStart=/home/shibbiry/go/bin/thyme track -o /home/shibbiry/logs/thyme.json
# if thyme has not finished in 5 seconds, consider it failed and kill it

philip-bl

class ShapyLinear(nn.Module):
    """Can model any affine function from the set of tensors of any (fixed) shape `in_shape` to
    the set of tensors of any (fixed) shape `out_shape`.
    
    In forward method the first modes of `inputs` are interpreted as indices of samples,
    then come the modes corresponding to `in_shape`. The affine function is applied to each sample."""
    def __init__(self, in_shape, out_shape):
        """:param in_shape: shape of one input sample
        :param out_shape: shape of one output sample"""
        super().__init__()

philip-bl

def set_random_seeds(seed_value=555, device='cuda:0'):
    '''source https://forums.fast.ai/t/solved-reproducibility-where-is-the-randomness-coming-in/31628/5.'''
    np.random.seed(seed_value)
    torch.manual_seed(seed_value)
    random.seed(seed_value)
    if device != 'cpu': 
        torch.cuda.manual_seed(seed_value)
        torch.cuda.manual_seed_all(seed_value)
        torch.backends.cudnn.deterministic = True
        torch.backends.cudnn.benchmark = False

philip-bl

def free_cuda_memory():
    import gc
    gc.collect()
    torch.cuda.empty_cache()

philip-bl