🐛 bug fix in choose
🎉 Any method with __show__ will act as a shortcut for the show function
✨ current_file_dir is a new function that return's the directory where the file exists
🐛 choose can optionally be silent (useful in funcs like rand)
🎉 ll method in P
🐛 tryy will Warn instead of print during an exception
✨ track2 can take optional info to display
🐛 fixed a bug where timer was asserting N > 0 instead of N>=0
✨ printfile will act like cat for linux
✨ choose will let you know what was chosen
✨ Timer can accept N=None and it will simply time the current activities without any estimate
✨ tryy is pickleable (helps with multiprocessing)
🎉 Add profiler
🎉 environment variable AD_SHOW_FULL_STRINGS can be set to non-empty to show full strings in summary
🎉 better integration for unknown datatypes in AD.summary
🎉 Quality of life improvements in logging
🎉 Add support for dataclass in AD
🎉 DeepDiff integrated into AD. Now you can call x.diff(y) where x is AD and y is ADable
🎉 AD.d (no brackets, d is a property) to convert the object to dictionary
✨ read_json can also do json.loads if input is not a file path
✨ json is loaded by default on import *
✨ Made all imports lazy for a faster import experience
✨ set AD_SHOW_TAB_STOPS env variable to anything to print with tab-stops (only recommended to see on REPL. Default is good for storing files)
✨ crop_from_bb has padding (px,py,pX,pY)
🐞 minor bugfix in AD
🎉 AD has a fetch2 which preserves json structure while fetching all the leafs of the same name
🎉 AD has a flatten which flattens all nested jsons into flat dictionary with '.' combined key names
🎉 AD has a flatten_and_make_dataframe which creates one column for each nesting
✨ tryy will store all its errors and can be accessed by calling <func>.error_summary()
🎉 torch snippets as a new cli function called ts, thanks to the wonderful Typer. Try
ts time
ts zen-of python
ts zen-of c++
ts --help
✨ slightly better error printing in tryy
🎉 tryy a new decorator to wrap try except with optional default return as well as print stack trace if need be
✨ Timer can given time-remaning approximation based on both instantaneous or average speed
✨ logging_mode functions can be used as both context managers and decorators
✨ better formatting for now()
✨ show auto generates a title if possible
✨ show function includes a framecount parameter to specify which frame of the stack it should search for the title. See its usage in show_big_dataframe
✨ yolo_2_df and df_2_yolo functions in adapters
✨ AD let's you know if a collection is list, set or tuple
✨ NEW DEFAULT: use info column to show text else text
🐞 fix a bug where P misbehaves in AD when it doesn't exist
🐞 fix a bug in show tables
🐞 logger_mode context will gracefully exit
✨ to is compatible with transformers.BatchFeature
⏮️ Revert back P as the code is incompatible with py312
🆕 parseing a .ini file will add a META key by default which has file and name information. This header can be used in the file variables without creating META to begin with
🆕 parse will merge a base config if a META.base key is found. Works recursively on base files (base of base etc...)
🆕 P can tab complete existing file path name and access that file as an attribute
💡 Did you know parse can perform arithmetic?
✨ registry and parse are imported by default
✨ jitter can work on pd.DataFrame directly
🔨 bbs2df clips negative coordinates
✨ Using 🏷️ to denote data-type in AD
🐞 AD2 is AD now
🐞 AD preserves dict order during serialization
✨ warn_on_fail decorator gives you a free try catch
✨ showing a dataframe of the correct format inside and outside jupyter environments
🐞 AD has improved get, update, map and new write_config, trymap and __json__
✨ P has two new properties .sz for size of file and .tree to print the tree
🐞 AD hotfix
🎉 store_scrap is a new way to store on disk and show jupyter cell outputs in other notebooks.
Best for presenting complex analyses without worrying about running time-consuming notebooks
🎉 Add support for P in AD summary and write_json
🎉 __json__ supports custom objects' serializability
🎉 write_json is compatible with above feature
🎉 AD_MAX_ITEMS if given as -1 will change it to 1000
🐞 iou will parse input dataframes more gracefully
🐞 AD minor bug fix
🎉 tree has a better default
🎉 New functions folder_structure_to_dict and folder_structure_to_json in paths
🎉 Add jitter (int) argument to show so that bounding boxes can be a bit jittered
🎉 Add support for changing spinner in notify_waiting
🎉 dumpdill can print a custom message (see store_scrap in paths.py)
🎉 AD __contains__ can do a nested in 'x.y.z' in AD(x={'y': {'z': 10}}) == True
☠️ Stop using rich's print and revert back to builtin print 🐞 Decouple AD and torch 🎉 Add a new chart - spider / radar 🎉 Add scp client with download upload functionality
🧹 change code to remove future warnings in text_utils
🐞 AD string summary was buggy
🐞 all write modes are 'a' by default to avoid accidental overwriting 🧹 AD writes better string summary (support for multiline)
🧹 print_module_ios_for has better targeted functionality where you need to give submodules name
🎉 clean_gpu_mem and get_latest_checkpoint functions in torch_loader
🧹 minor bugfix in AD
🐞 minor bugfix in AD2 where data keyword misbehaves
🧹 video has better size functionality
🎉 if all is given in print_ios_for_module, all modules are printed
🎉 AD2.dict is an alias for AD2.to_dict
🎉 better AD2.summary for pandas dataframes and AD2.summary respects max_items for keys as well
🎉 new alias pd.read_pqt for pd.read_parquet
🐞 wrap tree into python
🐞 read loads color image by default
🧹 video utils are present in torch_snippets.video
🧹 Back to min python version 3.7
🎉 print_folder_summary
🧹 import AD from markup2 by default
🎉 pandas dataframe summary in AD.summary
🐞 print_shapes_hook will gracefully fail
🐞 attach will add hook to the input module as well (not just the children)
🐞 minor change in print_shapes_hook
🎉 print_module_io_for automates attaching and detaching hooks 🎉 AD2 avoids rich printing
🐞 attach_hooks will accept any custom hook
🎉 Make markup2.AD.__repr__ the summary
🎉 Expose markup2.AD as AD2
🎉 Make icecream a requirement
🎉 Min python is 3.8
🎉 New IO hooks system in torch_snippets.trainer.hooks
🎉 Updated markup2.AD.summary and add print_summary methods
🎉 Experimental AD in torch_snippets.markup2 that infers variable names E.g. - (p=10; AD(p) == {'p': 10})
🐞 isin will not add +1 (useful for both absolute and relative boxes now)
🐞 write_json will support numpy, torch and AttrDict
🎉 add find_address to AttrDict that can return all path locations for a specific key
🎉 add summary to AttrDict that can give an outline of the dictionary
🎉 add write_summary to AttrDict that writes the summary to a textfile
🎉 show can now show bb colors {"r": (255, 0, 0), "g": (0, 255, 0), "b": (0, 0, 255), "y": (255, 0, 255)} if df has column called color
🎉 AD is an alias for AttrDict
🎉 AD can directly consume kwargs
🧹 import only important functins from dates.py
🎉 add backup_all_notebooks that backs up every notebook present in a specific folder
🎉 reset_logger can disable stdout logging if needed, using disable_stdout=True kwarg (False by default)
common_items will take a list of folders and return common stems from the folders
images will show a black border when grid is True
Info, Debug, Warn and Excep will format ouputs separated by a ; when args are passed
notify_waiting is a new function that letting you know some process is running for an unknown amount of time
optional delay during shutdown_current_notebook
Info, Debug, Warn and Excep will all have X_mode and in_X_mode functions much like in_debug_mode and debug_mode
__init__ will auto pull from logger now
Better non-linear Timer (and Report and track2)
🧹 Info, Debug, Warn and Excep will accept args (instead of a single arg)
🧹 show will show h4 headers instead of h2 for dataframe titles
🧹 phasify loads by default
🧹 show_big_dataframe can show more rows
🎉 add a new submodule trainer.hooks
🧹 show delegated kwargs to plt.imshow for a better readme
🎉 batchify can batchify multiple containers at once
🎉 cat_with_padding new function in torch_loader
🧹 L is json compatible
🐞 BB will not decide if something is relative/absolute
🎉 __contains__ for config
🎉 to works on AttrDict
👶🏼 track2 is a better version of track uses corouties
👶🏼 debug_mode temporarily activates DEBUG mode on
👶🏼 if in_debug_mode(): lets you know if DEBUG mode is on
🧹 reset_logger can accept lowercase levels also
🧹 dumpdill will return a Path after dumping
bugfix in loader.show
add today function to dates
add are_equal_dates to dates
add dpi option to pdf
bugfix in attrdict.map
All notebooks are formatted with black
parse can parse python expressions
Add DeepLearningConfig class that can be used to load model hyperparameters
Add GenericConfig class that can be used to load generic (such as training, evaluation) hyperparameters
Add date utilities
patch_to, Timer, timeit, io are loaded by default
lovely_tensors is optional
Add phasify function to loader
attrdict can deserialize "L"
showcan render a dataframe with a titleshowcan accept a csv file as input (no need to load it and send)backup_this_notebookwill back up asbackups/file/file__0000.htmlinstead ofbackups/file/0000.htmlfor easier sharability- module loads
decoratorsby default (io,timeit,check_kwargs_not_none) ishowis less opinionatedshutdown_this_notebookis a new function
override_previous_backup should not trigger when there's no backup to begin with instead of showing markdown objects using display, directly show HTML objects so that the text is preserved on reopen h2 in Backup instead of h1