Skip to main content

One line functions for common tasks

Project description

Utilities for simple needs

# Time it
from torch_snippets import *
CPU times: user 1.57 s, sys: 1.59 s, total: 3.16 s
Wall time: 731 ms

Below we are trying to extract the __all__ list from all Python files of the torch_snippets directory.
Through the code, you can already see some of the elements of torch-snippets in action.

import ast

os.environ[
    "AD_MAX_ITEMS"
] = (  # os is already imported by torch_snippets, along with many other useful libraries
    "1000"  # Set the maximum number of items to display in the AD object
)


@tryy  # This is a decorator that catches exceptions
def extract_all_list(file_path):
    file = readfile(file_path, silent=True)  # Read the file
    tree = ast.parse(file, filename=file_path)

    for node in tree.body:
        if isinstance(node, ast.Assign):
            for target in node.targets:
                if isinstance(target, ast.Name) and target.id == "__all__":
                    if isinstance(node.value, ast.List):
                        all_list = [
                            elt.value
                            for elt in node.value.elts
                            if isinstance(elt, ast.Constant)
                        ]
                        return all_list
    return None


def print_all_lists_in_directory(directory):
    dir = P(directory)  # Create a pathlib.Path object
    for f in dir.ls():  # Iterate over all files in the directory
        if f.extn == "py" and f.stem not in [
            "__init__",
            "_nbdev",
        ]:  # If it's a Python file and not __init__.py
            all_list = extract_all_list(f)
            if all_list is not None and len(all_list) > 0:
                h2(f.stem)  # Print the name of the file as a heading in jupyter
                print(
                    AD({"items": all_list})
                )  # AD is an intelligent dictionary that can display itself nicely
print(P().resolve())
/Users/yeshwanth/Code/Personal/torch_snippets/nbs
# Specify the directory containing the Python files
directory_path = "../torch_snippets"
print_all_lists_in_directory(directory_path)

misc

```↯ AttrDict ↯
items[]
  0 - Timer (🏷️ str)
  1 - track2 (🏷️ str)
  2 - summarize_input (🏷️ str)
  3 - timeit (🏷️ str)
  4 - io (🏷️ str)
  5 - tryy (🏷️ str)

```

load_defaults

```↯ AttrDict ↯
items[]
  0 - ifnone (🏷️ str)
  1 - exists (🏷️ str)
  2 - loadifexists (🏷️ str)

```

text_utils

```↯ AttrDict ↯
items[]
  0 - textify (🏷️ str)
  1 - find_lines (🏷️ str)
  2 - find_blocks (🏷️ str)
  3 - find_substring (🏷️ str)
  4 - get_line_data_from_word_data (🏷️ str)
  5 - edit_distance_path (🏷️ str)
  6 - group_blocks (🏷️ str)

```

paths

```↯ AttrDict ↯
items[]
  0 - valid_methods (🏷️ str)
  1 - P (🏷️ str)
  2 - ls (🏷️ str)
  3 - print_folder_summary (🏷️ str)
  4 - dill (🏷️ str)
  5 - input_to_str (🏷️ str)
  6 - output_to_path (🏷️ str)
  7 - process_f (🏷️ str)
  8 - get_fs (🏷️ str)
  9 - P0 (🏷️ str)
  10 - stem (🏷️ str)
  11 - stems (🏷️ str)
  12 - extn (🏷️ str)
  13 - remove_file (🏷️ str)
  14 - isdir (🏷️ str)
  15 - makedir (🏷️ str)
  16 - fname (🏷️ str)
  17 - fname2 (🏷️ str)
  18 - parent (🏷️ str)
  19 - Glob (🏷️ str)
  20 - find (🏷️ str)
  21 - zip_files (🏷️ str)
  22 - unzip_file (🏷️ str)
  23 - list_zip (🏷️ str)
  24 - md5 (🏷️ str)
  25 - remove_duplicates (🏷️ str)
  26 - common_items (🏷️ str)
  27 - folder_summary (🏷️ str)
  28 - readlines (🏷️ str)
  29 - readfile (🏷️ str)
  30 - writelines (🏷️ str)
  31 - tree (🏷️ str)
  32 - folder_structure_to_dict (🏷️ str)
  33 - folder_structure_to_json (🏷️ str)
  34 - rename_batch (🏷️ str)
  35 - dumpdill (🏷️ str)
  36 - loaddill (🏷️ str)

```

charts

```↯ AttrDict ↯
items[]
  0 - alt (🏷️ str)
  1 - Chart (🏷️ str)
  2 - CM (🏷️ str)
  3 - radar (🏷️ str)
  4 - confusion_matrix (🏷️ str)
  5 - spider (🏷️ str)
  6 - upsetaltair_top_level_configuration (🏷️ str)
  7 - UpSetAltair (🏷️ str)

```

pdf_loader

```↯ AttrDict ↯
items[]
  0 - PDF (🏷️ str)
  1 - dump_pdf_images (🏷️ str)
  2 - preview_pdf (🏷️ str)

```

interactive_show

```↯ AttrDict ↯
items[]
  0 - COLORS (🏷️ str)
  1 - to_networkx (🏷️ str)
  2 - plot_image (🏷️ str)
  3 - plot_graph (🏷️ str)
  4 - tonp (🏷️ str)
  5 - tolist (🏷️ str)
  6 - convert_to_nx (🏷️ str)
  7 - viz2 (🏷️ str)
  8 - df2graph_nodes (🏷️ str)
  9 - ishow (🏷️ str)

```

registry

```↯ AttrDict ↯
items[]
  0 - Config (🏷️ str)
  1 - AttrDict (🏷️ str)
  2 - registry (🏷️ str)
  3 - tryeval (🏷️ str)
  4 - parse_base (🏷️ str)
  5 - parse (🏷️ str)
  6 - parse_and_resolve (🏷️ str)
  7 - parse_string (🏷️ str)

```

markup2

```↯ AttrDict ↯
items[]
  0 - AD (🏷️ str)
  1 - Config (🏷️ str)
  2 - isnamedtupleinstance (🏷️ str)
  3 - unpack (🏷️ str)
  4 - AttrDict (🏷️ str)
  5 - pretty_json (🏷️ str)
  6 - read_json (🏷️ str)
  7 - write_json (🏷️ str)
  8 - write_jsonl (🏷️ str)
  9 - read_jsonl (🏷️ str)
  10 - read_yaml (🏷️ str)
  11 - write_yaml (🏷️ str)
  12 - read_xml (🏷️ str)
  13 - write_xml (🏷️ str)

```

inspector

```↯ AttrDict ↯
items[]
  0 - inspect (🏷️ str)

```

torch_loader

```↯ AttrDict ↯
items[]
  0 - torch (🏷️ str)
  1 - th (🏷️ str)
  2 - torchvision (🏷️ str)
  3 - T (🏷️ str)
  4 - transforms (🏷️ str)
  5 - nn (🏷️ str)
  6 - np (🏷️ str)
  7 - F (🏷️ str)
  8 - Dataset (🏷️ str)
  9 - DataLoader (🏷️ str)
  10 - optim (🏷️ str)
  11 - Report (🏷️ str)
  12 - Reshape (🏷️ str)
  13 - Permute (🏷️ str)
  14 - device (🏷️ str)
  15 - save_torch_model_weights_from (🏷️ str)
  16 - load_torch_model_weights_to (🏷️ str)
  17 - detach (🏷️ str)
  18 - cat_with_padding (🏷️ str)

```

logger

```↯ AttrDict ↯
items[]
  0 - console (🏷️ str)
  1 - reset_logger_width (🏷️ str)
  2 - logger (🏷️ str)
  3 - Trace (🏷️ str)
  4 - Debug (🏷️ str)
  5 - Info (🏷️ str)
  6 - Warn (🏷️ str)
  7 - Excep (🏷️ str)
  8 - warn_mode (🏷️ str)
  9 - info_mode (🏷️ str)
  10 - debug_mode (🏷️ str)
  11 - trace_mode (🏷️ str)
  12 - excep_mode (🏷️ str)
  13 - in_warn_mode (🏷️ str)
  14 - in_info_mode (🏷️ str)
  15 - in_debug_mode (🏷️ str)
  16 - in_trace_mode (🏷️ str)
  17 - in_excep_mode (🏷️ str)
  18 - frames (🏷️ str)
  19 - get_console (🏷️ str)
  20 - reset_logger (🏷️ str)
  21 - get_logger_level (🏷️ str)
  22 - logger_mode (🏷️ str)
  23 - in_logger_mode (🏷️ str)
  24 - notify_waiting (🏷️ str)

```

markup

```↯ AttrDict ↯
items[]
  0 - AttrDict (🏷️ str)
  1 - json (🏷️ str)
  2 - Config (🏷️ str)
  3 - isnamedtupleinstance (🏷️ str)
  4 - unpack (🏷️ str)
  5 - hash_tensor (🏷️ str)
  6 - hash_pandas_dataframe (🏷️ str)
  7 - AttrDictDeprecated (🏷️ str)
  8 - decompose (🏷️ str)
  9 - pretty_json (🏷️ str)
  10 - read_json (🏷️ str)
  11 - write_json (🏷️ str)
  12 - write_jsonl (🏷️ str)
  13 - read_jsonl (🏷️ str)
  14 - read_yaml (🏷️ str)
  15 - write_yaml (🏷️ str)
  16 - read_xml (🏷️ str)
  17 - write_xml (🏷️ str)

```

sklegos

```↯ AttrDict ↯
items[]
  0 - ColumnSelector (🏷️ str)
  1 - GroupedPredictor (🏷️ str)
  2 - EstimatorTransformer (🏷️ str)
  3 - train_test_split (🏷️ str)
  4 - MakeFrame (🏷️ str)
  5 - ImputeMissingValues (🏷️ str)
  6 - LambdaTransformer (🏷️ str)
  7 - Cat2Num (🏷️ str)
  8 - SplitDateColumn (🏷️ str)

```

ipython

```↯ AttrDict ↯
items[]
  0 - save_notebook (🏷️ str)
  1 - backup_this_notebook (🏷️ str)
  2 - backup_all_notebooks (🏷️ str)
  3 - backup_folders_of_nbs (🏷️ str)
  4 - display_dfs_side_by_side (🏷️ str)
  5 - show_big_dataframe (🏷️ str)
  6 - h1 (🏷️ str)
  7 - h2 (🏷️ str)
  8 - h3 (🏷️ str)
  9 - h4 (🏷️ str)
  10 - h5 (🏷️ str)
  11 - h6 (🏷️ str)
  12 - store_scrap (🏷️ str)
  13 - shutdown_current_notebook (🏷️ str)

```

../torch_snippets/loader.py:532: SyntaxWarning: invalid escape sequence '\$'
  puttext(ax, text.replace("$", "\$"), tuple(bbs[ix][:2]), size=text_sz)

loader

```↯ AttrDict ↯
items[]
  0 - B (🏷️ str)
  1 - Blank (🏷️ str)
  2 - batchify (🏷️ str)
  3 - C (🏷️ str)
  4 - choose (🏷️ str)
  5 - common (🏷️ str)
  6 - crop_from_bb (🏷️ str)
  7 - diff (🏷️ str)
  8 - E (🏷️ str)
  9 - flatten (🏷️ str)
  10 - Image (🏷️ str)
  11 - jitter (🏷️ str)
  12 - L (🏷️ str)
  13 - lzip (🏷️ str)
  14 - line (🏷️ str)
  15 - lines (🏷️ str)
  16 - to_absolute (🏷️ str)
  17 - to_relative (🏷️ str)
  18 - enlarge_bbs (🏷️ str)
  19 - shrink_bbs (🏷️ str)
  20 - logger (🏷️ str)
  21 - np (🏷️ str)
  22 - now (🏷️ str)
  23 - nunique (🏷️ str)
  24 - os (🏷️ str)
  25 - pad (🏷️ str)
  26 - pd (🏷️ str)
  27 - pdfilter (🏷️ str)
  28 - pdb (🏷️ str)
  29 - PIL (🏷️ str)
  30 - print (🏷️ str)
  31 - puttext (🏷️ str)
  32 - randint (🏷️ str)
  33 - rand (🏷️ str)
  34 - re (🏷️ str)
  35 - read (🏷️ str)
  36 - readPIL (🏷️ str)
  37 - rect (🏷️ str)
  38 - resize (🏷️ str)
  39 - rotate (🏷️ str)
  40 - see (🏷️ str)
  41 - show (🏷️ str)
  42 - store_attr (🏷️ str)
  43 - subplots (🏷️ str)
  44 - sys (🏷️ str)
  45 - toss (🏷️ str)
  46 - track (🏷️ str)
  47 - tqdm (🏷️ str)
  48 - Tqdm (🏷️ str)
  49 - trange (🏷️ str)
  50 - unique (🏷️ str)
  51 - uint (🏷️ str)
  52 - write (🏷️ str)
  53 - BB (🏷️ str)
  54 - bbfy (🏷️ str)
  55 - xywh2xyXY (🏷️ str)
  56 - df2bbs (🏷️ str)
  57 - bbs2df (🏷️ str)
  58 - Info (🏷️ str)
  59 - Warn (🏷️ str)
  60 - Debug (🏷️ str)
  61 - Excep (🏷️ str)
  62 - reset_logger (🏷️ str)
  63 - get_logger_level (🏷️ str)
  64 - in_debug_mode (🏷️ str)
  65 - debug_mode (🏷️ str)
  66 - typedispatch (🏷️ str)
  67 - defaultdict (🏷️ str)
  68 - Counter (🏷️ str)
  69 - dcopy (🏷️ str)
  70 - patch_to (🏷️ str)
  71 - split (🏷️ str)
  72 - train_test_split (🏷️ str)
  73 - init_plt (🏷️ str)
  74 - init_cv2 (🏷️ str)

```

imgaug_loader

```↯ AttrDict ↯
items[]
  0 - do (🏷️ str)
  1 - bw (🏷️ str)
  2 - rotate (🏷️ str)
  3 - pad (🏷️ str)
  4 - get_size (🏷️ str)
  5 - rescale (🏷️ str)
  6 - crop (🏷️ str)
  7 - imgaugbbs2bbs (🏷️ str)
  8 - bbs2imgaugbbs (🏷️ str)

```

dates

```↯ AttrDict ↯
items[]
  0 - make_uniform_date_format (🏷️ str)
  1 - ALL_DATE_FORMATS (🏷️ str)
  2 - are_dates_equal (🏷️ str)
  3 - today (🏷️ str)

```

profiler

```↯ AttrDict ↯
items[]
  0 - time_profiler (🏷️ str)

```

bokeh_loader

```↯ AttrDict ↯
items[]
  0 - parse_sz (🏷️ str)
  1 - get_bplot (🏷️ str)

```

bb_utils

```↯ AttrDict ↯
items[]
  0 - randint (🏷️ str)
  1 - BB (🏷️ str)
  2 - df2bbs (🏷️ str)
  3 - bbs2df (🏷️ str)
  4 - bbfy (🏷️ str)
  5 - jitter (🏷️ str)
  6 - compute_eps (🏷️ str)
  7 - enlarge_bbs (🏷️ str)
  8 - shrink_bbs (🏷️ str)
  9 - iou (🏷️ str)
  10 - compute_distance_matrix (🏷️ str)
  11 - compute_distances (🏷️ str)
  12 - split_bb_to_xyXY (🏷️ str)
  13 - combine_xyXY_to_bb (🏷️ str)
  14 - is_absolute (🏷️ str)
  15 - is_relative (🏷️ str)
  16 - to_relative (🏷️ str)
  17 - to_absolute (🏷️ str)
  18 - merge_by_bb (🏷️ str)
  19 - isin (🏷️ str)

```

adapters

```↯ AttrDict ↯
items[]
  0 - np_2_b64 (🏷️ str)
  1 - b64_2_np (🏷️ str)
  2 - b64_2_file (🏷️ str)
  3 - bytes_2_file (🏷️ str)
  4 - file_2_bytes (🏷️ str)
  5 - csvs_2_cvat (🏷️ str)
  6 - cvat_2_csvs (🏷️ str)
  7 - df_2_yolo (🏷️ str)
  8 - yolo_2_df (🏷️ str)

```

decorators

```↯ AttrDict ↯
items[]
  0 - format (🏷️ str)
  1 - warn_on_fail (🏷️ str)
  2 - timeit (🏷️ str)
  3 - io (🏷️ str)
  4 - check_kwargs_not_none (🏷️ str)

```

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

torch_snippets-0.554.tar.gz (96.0 kB view details)

Uploaded Source

Built Distribution

torch_snippets-0.554-py3-none-any.whl (108.1 kB view details)

Uploaded Python 3

File details

Details for the file torch_snippets-0.554.tar.gz.

File metadata

  • Download URL: torch_snippets-0.554.tar.gz
  • Upload date:
  • Size: 96.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.12.10

File hashes

Hashes for torch_snippets-0.554.tar.gz
Algorithm Hash digest
SHA256 ac085fbc0b5e8927bcbe8c648d4b736da729f5ecd839a69125ea599f28dc37d6
MD5 e170bf2c6229d340060654bb5b929176
BLAKE2b-256 6dfdb238833bc95713a96747f8e8e7ff22a69449a5f033e79e410ab33be8b70e

See more details on using hashes here.

File details

Details for the file torch_snippets-0.554-py3-none-any.whl.

File metadata

  • Download URL: torch_snippets-0.554-py3-none-any.whl
  • Upload date:
  • Size: 108.1 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.12.10

File hashes

Hashes for torch_snippets-0.554-py3-none-any.whl
Algorithm Hash digest
SHA256 bd765101eeb0b2f9bffbbcfcb2a0ec49e74ea95c03205d5345ba898e211d831a
MD5 9ec60646ec8132d389ce19eeab2a63ad
BLAKE2b-256 99dfae3ec1232d06b323bc8f30e0997106808e1aab92c2348cb0936b7bade6b4

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page