U
|
©r_¤? ã @ sÊ d dl Z d dlZd dlZd dlZd dlZd dlmZ d dlmZ d dl Z ddl
|
m Z ddl m Z ddd d
|
d gZg Zd d ZG d d dZG dd deZG dd d eZG dd
|
d
|
eZG dd d ZdS )é N)Ú defaultdict)Úcontextmanageré )Ú PathManager)Ú HistoryBufferÚget_event_storageÚ
|
JSONWriterÚTensorboardXWriterÚCommonMetricPrinterÚ EventStoragec C s t tstdtd S )z¢
|
Returns:
|
The :class:`EventStorage` object that's currently being used.
|
Throws an error if no :class:`EventStorage` is currently enabled.
|
zOget_event_storage() has to be called inside a 'with EventStorage(...)' context!éÿÿÿÿ)ÚlenÚ_CURRENT_STORAGE_STACKÚAssertionError© r r ú./utils/events.pyr s ÿþc @ s e Zd ZdZdd Zdd ZdS )Ú EventWriterz`
|
Base class for writers that obtain events from :class:`EventStorage` and process them.
|
c C s t d S ©N)ÚNotImplementedError©Úselfr r r Úwrite) s zEventWriter.writec C s d S r r r r r r Úclose, s zEventWriter.closeN)Ú__name__Ú
|
__module__Ú __qualname__Ú__doc__r r r r r r r $ s r c @ s* e Zd ZdZd
|
ddZdd Zdd Zd S ) r a
|
Write scalars to a json file.
|
It saves scalars as one json per line (instead of a big json) for easy parsing.
|
Examples parsing such a json file:
|
::
|
$ cat metrics.json | jq -s '.[0:2]'
|
[
|
{
|
"data_time": 0.008433341979980469,
|
"iteration": 20,
|
"loss": 1.9228371381759644,
|
"loss_box_reg": 0.050025828182697296,
|
"loss_classifier": 0.5316952466964722,
|
"loss_mask": 0.7236229181289673,
|
"loss_rpn_box": 0.0856662318110466,
|
"loss_rpn_cls": 0.48198649287223816,
|
"lr": 0.007173333333333333,
|
"time": 0.25401854515075684
|
},
|
{
|
"data_time": 0.007216215133666992,
|
"iteration": 40,
|
"loss": 1.282649278640747,
|
"loss_box_reg": 0.06222952902317047,
|
"loss_classifier": 0.30682939291000366,
|
"loss_mask": 0.6970193982124329,
|
"loss_rpn_box": 0.038663312792778015,
|
"loss_rpn_cls": 0.1471673548221588,
|
"lr": 0.007706666666666667,
|
"time": 0.2490077018737793
|
}
|
]
|
$ cat metrics.json | jq '.loss_mask'
|
0.7126231789588928
|
0.689423680305481
|
0.6776131987571716
|
...
|
é c C s t |d¡| _|| _d| _dS )zý
|
Args:
|
json_file (str): path to the json file. New data will be appended if the file exists.
|
window_size (int): the window size of median smoothing for the scalars whose
|
`smoothing_hint` are True.
|
Úar N)r ÚopenÚ _file_handleÚ _window_sizeÚ _last_write)r Z json_fileÚ window_sizer r r Ú__init__X s zJSONWriter.__init__c C sÆ t }tt}| | j¡ ¡ D ]$\}\}}|| jkr6q ||| |< q t| ¡ }t || _| ¡ D ]*\}}||d< | j
|
t j |ddd ¡ qb| j
|
¡ zt | j
|
¡ ¡ W n tk
|
rÀ Y nX d S )NÚ iterationT)Z sort_keysÚ
|
)r r ÚdictÚlatest_with_smoothing_hintr! Úitemsr" ÚsortedÚkeysÚmaxr r ÚjsonÚdumpsÚflushÚosÚfsyncÚfilenoÚAttributeError) r ÚstorageÚto_saveÚkÚvÚiterZ all_itersÚitrZscalars_per_iterr r r r c s
|
|
|
zJSONWriter.writec C s | j ¡ d S r )r r r r r r r x s zJSONWriter.closeN)r )r r r r r$ r r r r r r r 0 s '
|
c @ s2 e Zd ZdZd eedddZdd Zdd Zd
|
S ) r z2
|
Write all scalars to a tensorboard file.
|
r )Úlog_dirr# c K s* || _ ddlm} ||f|| _d| _dS )a
|
Args:
|
log_dir (str): the directory to save the output events
|
window_size (int): the scalars will be median-smoothed by this window size
|
kwargs: other arguments passed to `torch.utils.tensorboard.SummaryWriter(...)`
|
r )Ú SummaryWriterr N)r! Ztorch.utils.tensorboardr; Ú_writerr" )r r: r# Úkwargsr; r r r r$ s zTensorboardXWriter.__init__c
|
C sÀ t }| j}| | j¡ ¡ D ]0\}\}}|| jkr | j |||¡ t||}q || _t|j dkr|j D ]\}}}| j
|
|||¡ qh| ¡ t|j dkr¼|j D ]} | jj f | q | ¡ d S )Nr )r r" r( r! r) r< Ú
|
add_scalarr, r Ú _vis_dataZ add_imageÚ clear_imagesÚ _histogramsZadd_histogram_rawÚclear_histograms)
|
r r4 Znew_last_writer6 r7 r8 Úimg_nameÚimgÚstep_numÚparamsr r r r s
|
|
zTensorboardXWriter.writec C s t | dr| j ¡ d S )Nr< )Úhasattrr< r r r r r r ¨ s
|
zTensorboardXWriter.closeN)r ) r r r r ÚstrÚintr$ r r r r r r r | s c @ s e Zd ZdZdd Zdd ZdS )r
|
aU
|
Print **common** metrics to the terminal, including
|
iteration time, ETA, memory, all losses, and the learning rate.
|
It also applies smoothing using a window of 20 elements.
|
It's meant to print common metrics in common ways.
|
To print something in more customized ways, please implement a similar printer by yourself.
|
c C s t t¡| _|| _d| _dS )z
|
Args:
|
max_iter (int): the maximum number of iterations to train.
|
Used to compute ETA.
|
N)ÚloggingÚ getLoggerr ÚloggerÚ _max_iterr" )r Úmax_iterr r r r$ ¶ s z CommonMetricPrinter.__init__c
|
C sæ t }|j}z| d¡ d¡}W n tk
|
r8 d }Y nX d }zP| d¡ ¡ }| d¡ d¡| j| }|jd|dd t t
|
j t |d}W nt tk
|
r d }| j d k rðt ¡ | j d || j d
|
}|| j| }t t
|
j t |d}|t ¡ f| _ Y nX zd | d ¡ ¡ ¡}W n tk
|
r6 d }Y nX tj ¡ rXtj ¡ d d } nd } | j dj|rxd| dnd|d dd | ¡ ¡ D ¡|d k rªd |¡nd|d k rÀd |¡nd|| d k rØd | ¡ndd¡ d S )NÚ data_timer Útimeéè Ú eta_secondsF©Úsmoothing_hint)Úsecondsr r z{:.2e}ÚlrzN/Ag @zA {eta}iter: {iter} {losses} {time}{data_time}lr: {lr} {memory}zeta: z Ú c S s* g | ]"\}}d |krd || d¡¡qS )Úlossz
|
{}: {:.4g}r )ÚformatÚmedian)Ú.0r6 r7 r r r Ú
|
<listcomp>ì s þz-CommonMetricPrinter.write.<locals>.<listcomp>ztime: {:.4f} zdata_time: {:.4f} zmax_mem: {:.0f}M)Úetar8 ÚlossesrP rO rV Úmemory)r r8 ÚhistoryÚavgÚKeyErrorZ
|
global_avgrZ rM Ú
|
put_scalarrH ÚdatetimeÚ timedeltarI r" rP Ú perf_counterrY ÚlatestÚtorchÚcudaÚ is_availableÚmax_memory_allocatedrL ÚinfoÚjoinÚ historiesr) )
|
r r4 r% rO Z
|
eta_stringZ iter_timerR Zestimate_iter_timerV Z
|
max_mem_mbr r r r À sV
|
|
ÿ
|
|
þÿóÿzCommonMetricPrinter.writeN)r r r r r$ r r r r r r
|
s
|
c @ sº e Zd ZdZd,ddZdd Zd-dd Zdd
|
d d Zd.ddZdd Z dd Z
|
dd Z d/ddZ dd Z dd Zed d Zed d Zd!d" Zd#d$ Zed%d& Zd'd( Zd)d* Zd+S )0r z§
|
The user-facing class that provides metric storage functionalities.
|
In the future we may add support for storing / logging other types of data if needed.
|
r c C s2 t t| _i | _i | _|| _d| _g | _g | _dS )zX
|
Args:
|
start_iter (int): the iteration number to start with
|
rW N) r r Ú_historyÚ_smoothing_hintsÚ_latest_scalarsÚ_iterÚ_current_prefixr? rA )r Z
|
start_iterr r r r$ s
|
zEventStorage.__init__c C s | j ||| jf¡ dS )a:
|
Add an `img_tensor` associated with `img_name`, to be shown on
|
tensorboard.
|
Args:
|
img_name (str): The name of the image to put into tensorboard.
|
img_tensor (torch.Tensor or numpy.array): An `uint8` or `float`
|
Tensor of shape `[channel, height, width]` where `channel` is
|
3. The image format should be RGB. The elements in img_tensor
|
can either have values in [0, 1] (float32) or [0, 255] (uint8).
|
The `img_tensor` will be visualized in tensorboard.
|
N)r? Úappendrr )r rC Z
|
img_tensorr r r Ú put_image s zEventStorage.put_imageTc C st | j | }| j| }t|}| || j¡ || jf| j|< | j |¡}|dk rf||ksptd |¡n
|
|| j|< dS )a
|
Add a scalar `value` to the `HistoryBuffer` associated with `name`.
|
Args:
|
smoothing_hint (bool): a 'hint' on whether this scalar is noisy and should be
|
smoothed when logged. The hint will be accessible through
|
:meth:`EventStorage.smoothing_hints`. A writer may ignore the hint
|
and apply custom smoothing rule.
|
It defaults to True because most scalars we save need to be smoothed to
|
provide any useful signal.
|
Nz2Scalar {} was put with a different smoothing_hint!)
|
rs ro ÚfloatÚupdaterr rq rp Úgetr rY )r ÚnameÚvaluerT r` Z existing_hintr r r rc s
|
|
ÿþzEventStorage.put_scalarrS c K s& | ¡ D ]\}}| j|||d qdS )z©
|
Put multiple scalars from keyword arguments.
|
Examples:
|
storage.put_scalars(loss=my_loss, accuracy=my_accuracy, smoothing_hint=True)
|
rS N)r) rc )r rT r= r6 r7 r r r Ú put_scalars4 s zEventStorage.put_scalarsrQ c C s | ¡ ¡ | ¡ ¡ }}tj||d}tj|||d tjd}t|||t|t |
|
¡ t t
|
|d ¡|dd
¡ | ¡ | j d }| j |¡ dS )a9
|
Create a histogram from a tensor.
|
Args:
|
hist_name (str): The name of the histogram to put into tensorboard.
|
hist_tensor (torch.Tensor): A Tensor of arbitrary shape to be converted
|
into a histogram.
|
bins (int): Number of histogram bins.
|
)Úbinsr )ÚstartÚendZstepsÚdtypeé N) ÚtagÚminr, ÚnumÚsumZ sum_squaresZ bucket_limitsZ bucket_countsZ global_step)r Úitemr, rh ÚhistcÚlinspaceÚfloat32r' r rv r Útolistrr rA rt ) r Z hist_nameZ hist_tensorr| Zht_minZht_maxZ hist_countsZ
|
hist_edgesZ hist_paramsr r r Ú put_histogram= s
|
÷ zEventStorage.put_histogramc C s( | j |d¡}|dkr$td |¡|S )zQ
|
Returns:
|
HistoryBuffer: the scalar history for name
|
Nz#No history metric available for {}!)ro rx rb rY )r ry Úretr r r r` Z s zEventStorage.historyc C s | j S )ze
|
Returns:
|
dict[name -> HistoryBuffer]: the HistoryBuffer for all scalars
|
)ro r r r r rn d s zEventStorage.historiesc C s | j S )z³
|
Returns:
|
dict[str -> (float, int)]: mapping from the name of each scalar to the most
|
recent value and the iteration number its added.
|
)rq r r r r rg k s zEventStorage.latestr c C sF i }| j ¡ D ]2\}\}}| j| r4| j| |¡n||f||< q|S )a'
|
Similar to :meth:`latest`, but the returned values
|
are either the un-smoothed original latest value,
|
or a median of the given window_size,
|
depend on whether the smoothing_hint is True.
|
This provides a default behavior that other writers can use.
|
)rq r) rp ro rZ )r r# Úresultr6 r7 r9 r r r r( s s þ
|
z'EventStorage.latest_with_smoothing_hintc C s | j S )z
|
Returns:
|
dict[name -> bool]: the user-provided hint on whether the scalar
|
is noisy and needs smoothing.
|
)rp r r r r Úsmoothing_hints s z EventStorage.smoothing_hintsc C s | j d7 _ dS )zþ
|
User should call this function at the beginning of each iteration, to
|
notify the storage of the start of a new iteration.
|
The storage will then be able to associate the new data with the
|
correct iteration number.
|
r N©rr r r r r Ústep s zEventStorage.stepc C s | j S r r r r r r r8 s zEventStorage.iterc C s | j S r r r r r r r% s zEventStorage.iterationc C s t | ¡ | S r )r rt r r r r Ú __enter__ s
|
zEventStorage.__enter__c C s t d | kstt ¡ d S )Nr )r r Úpop)r Úexc_typeÚexc_valÚexc_tbr r r Ú__exit__¡ s zEventStorage.__exit__c c s& | j }| d¡d | _ dV || _ dS )z
|
Yields:
|
A context within which all the events added to this storage
|
will be prefixed by the name scope.
|
ú/N)rs Úrstrip)r ry Z
|
old_prefixr r r Ú
|
name_scope¥ s zEventStorage.name_scopec C s
|
g | _ dS )z
|
Delete all the stored images for visualization. This should be called
|
after images are written to tensorboard.
|
N)r? r r r r r@ ± s zEventStorage.clear_imagesc C s
|
g | _ dS )z
|
Delete all the stored histograms for visualization.
|
This should be called after histograms are written to tensorboard.
|
N)rA r r r r rB ¸ s z EventStorage.clear_histogramsN)r )T)rQ )r )r r r r r$ ru rc r{ r r` rn rg r( r r Úpropertyr8 r% r r r r r@ rB r r r r r ú s,
|
|
|
|
|
|
|
|
)rd r- rJ r0 rP Ú collectionsr Ú
|
contextlibr rh Úfile_ior Zhistory_bufferr Ú__all__r r r r r r
|
r r r r r Ú<module> s, û L1M
|
