o iG@sddlZddlZddlZ     dddZ    ddd Z    dd d Zd d ZgdZgdZ de de ddddddddddddddfddZ dS)NFinputoutputc sl|dkr td|t|krtdtd|dg} d} | \} } t| ||dd}| dkrEt| || ddn ttdd | |}tt||t||}t|t|d |}tt| |}| |}|r}|t||kr|t||}fd d tj d| |D}|r|| || ||tkr | S|} q#) aMake batch set from json dictionary :param Dict[str, Dict[str, Any]] sorted_data: dictionary loaded from data.json :param int batch_size: batch size :param int max_length_in: maximum length of input to decide adaptive batch size :param int max_length_out: maximum length of output to decide adaptive batch size :param int min_batch_size: mininum batch size (for multi-gpu) :param bool shortest_first: Sort from batch with shortest samples to longest if true, otherwise reverse :param str ikey: key to access input (for ASR ikey="input", for TTS, MT ikey="output".) :param int iaxis: dimension to access input (for ASR, TTS iaxis=0, for MT iaxis="1".) :param str okey: key to access output (for ASR, MT okey="output". for TTS okey="input".) :param int oaxis: dimension to access output (for ASR, TTS, MT oaxis=0, reserved for future research, -1 means all axis.) :return: List[List[Tuple[str, dict]]] list of batches rzInvalid batch_size=z#utts(z) is less than min_batch_size(z).TshapecSst|ddS)Nrrint)xr Q/home/ubuntu/.local/lib/python3.10/site-packages/espnet/utils/training/batchfy.py8sz batchfy_by_seq..rcg|]}|qSr r .0i sorted_datar r Hz"batchfy_by_seq..) ValueErrorlenrmaxmapminreversenprandomrandintextendappend)r batch_size max_length_inmax_length_outmin_batch_sizeshortest_firstikeyiaxisokeyoaxis minibatchesstart_infoilenolenfactorbsend minibatchmodadditional_minibatchr rr batchfy_by_seqsF        r4cCs|dkr td|t|}t|dd|ddd}t|dd|ddd} tdtt|g} d} d} d} d}d}||kr| | |krt|| | d|ddd|}t|| | d|ddd| }||kr|}||| d}||kr| d7} n |dkrtd|d||kr| | |ksQt|| t|| }|| |}|r|| |d }t| ||kr|t| |}| t| kr| |d | || dd } n)| | | |dd || |d|d | |d<|d8}t| ||ks||krn|} | d7} qA|dkr.| d |} d d | D}ttt| d tt|dtt|ddttt |d| S)aMake variably sized batch set, which maximizes the number of bins up to `batch_bins`. :param Dict[str, Dict[str, Any]] sorted_data: dictionary loaded from data.json :param int batch_bins: Maximum frames of a batch :param int num_batches: # number of batches to use (for debug) :param int min_batch_size: minimum batch size (for multi-gpu) :param int test: Return only every `test` batches :param bool shortest_first: Sort from batch with shortest samples to longest if true, otherwise reverse :param str ikey: key to access input (for ASR ikey="input", for TTS ikey="output".) :param str okey: key to access output (for ASR okey="output". for TTS okey="input".) :return: List[Tuple[str, Dict[str, List[Dict[str, Any]]]] list of batches rzinvalid batch_bins=rr# utts: Tz$Can't fit one sample in batch_bins (): Please increase the valueNcSg|]}t|qSr rrrr r r rz"batchfy_by_bin.. batches containing from  to z samples (avg samples).) rrrloggingr+strrrrrrrmean)r batch_bins num_batchesr"r#r$r&lengthidimodimr(r)nb next_sizemax_olenr,r-r0batchrmissinglengthsr r r batchfy_by_binXs  ((       '     rOc Cs|dkr|dkr|dkrtdt|} g} d} d} | | kr7d} d}d}| | | krt|| | d|ddd}||krM|dkrMtd|dt|| | d|ddd}||kro|dkrotd|d|||kr|dkrtd|dt||}t||}|| d|kp|dk}|| d|kp|dk}||| d|kp|dk}|r|r|r| d7} nn| | | ks+t| | | } || | }|r|| |d}t| ||kr1|t| |}| t| kr| |d| || dd } n)| || |dd || |d|d | |d<|d8}t| ||ks| } | | ks|dkrB| d |} d d | D}t t t| d t t|d t t|ddt tt |d| S)aMake variable batch set, which maximizes the number of frames to max_batch_frame. :param Dict[str, Dict[str, Any]] sorteddata: dictionary loaded from data.json :param int max_frames_in: Maximum input frames of a batch :param int max_frames_out: Maximum output frames of a batch :param int max_frames_inout: Maximum input+output frames of a batch :param int num_batches: # number of batches to use (for debug) :param int min_batch_size: minimum batch size (for multi-gpu) :param int test: Return only every `test` batches :param bool shortest_first: Sort from batch with shortest samples to longest if true, otherwise reverse :param str ikey: key to access input (for ASR ikey="input", for TTS ikey="output".) :param str okey: key to access output (for ASR okey="output". for TTS okey="input".) :return: List[Tuple[str, Dict[str, List[Dict[str, Any]]]] list of batches rzbAt least, one of `--batch-frames-in`, `--batch-frames-out` or `--batch-frames-inout` should be > 0rrz+Can't fit one sample in --batch-frames-in (r6z,Can't fit one sample in --batch-frames-out (r7NcSr8r r9r:r r r rr;z$batchfy_by_frame..r<r=z samplesr>r?) rrrrrrrrr@r+rArrB)r max_frames_inmax_frames_outmax_frames_inoutrDr"r#r$r&rEr(r)r0rIrKmax_ilenr,r-in_okout_okinout_okrLrrMrNr r r batchfy_by_frames  $ $            6     rWc sddl}td||t|tdttg}d} tt||}||} |r:| t| |krb|t| |} fddt j d|| D} |r]| | | | | |tkrnn|}q%|dkr|d|}tdtt||S)Nrzuse shuffled batch.r5Tcr r r r rr r r1rz#batchfy_shuffle..# minibatches: ) rr@r+sampleitemsrrArrrrrr) datarr"rDr#rr(r)r0r1r2r3r rr batchfy_shuffle s8        r\)autoseqbinframe)rrshuffleinfr]c sj| tvrtd| dttvrtddtd|r2d}d}dkr+dn"dkr1dn| rId}d}dd|dksBJ|dksHJnd}d}| dkr}|dkrXd } n| dkr_d } n| dksk| dksk|dkrnd } ntd ttd | | d krdkrtdi}|D]\}}|||di|<qg}|D]d}dkrt |||||}| |qt |fdd| d}tdt t || d krt||||||||||d }| d krt|| ||||d}| d krt|| | |||||d}| |qt |dkr|d}nttj|}|dkr(|d|}tdt t ||S)aMake batch set from json dictionary if utts have "category" value, >>> data = {'utt1': {'category': 'A', 'input': ...}, ... 'utt2': {'category': 'B', 'input': ...}, ... 'utt3': {'category': 'B', 'input': ...}, ... 'utt4': {'category': 'A', 'input': ...}} >>> make_batchset(data, batchsize=2, ...) [[('utt1', ...), ('utt4', ...)], [('utt2', ...), ('utt3': ...)]] Note that if any utts doesn't have "category", perform as same as batchfy_by_{count} :param Dict[str, Dict[str, Any]] data: dictionary loaded from data.json :param int batch_size: maximum number of sequences in a minibatch. :param int batch_bins: maximum number of bins (frames x dim) in a minibatch. :param int batch_frames_in: maximum number of input frames in a minibatch. :param int batch_frames_out: maximum number of output frames in a minibatch. :param int batch_frames_out: maximum number of input+output frames in a minibatch. :param str count: strategy to count maximum size of batch. For choices, see espnet.asr.batchfy.BATCH_COUNT_CHOICES :param int max_length_in: maximum length of input to decide adaptive batch size :param int max_length_out: maximum length of output to decide adaptive batch size :param int num_batches: # number of batches to use (for debug) :param int min_batch_size: minimum batch size (for multi-gpu) :param bool shortest_first: Sort from batch with shortest samples to longest if true, otherwise reverse :param str batch_sort_key: how to sort data before creating minibatches ["input", "output", "shuffle"] :param bool swap_io: if True, use "input" as output and "output" as input in `data` dict :param bool mt: if True, use 0-axis of "output" as output and 1-axis of "output" as input in `data` dict :param int iaxis: dimension to access input (for ASR, TTS iaxis=0, for MT iaxis="1".) :param int oaxis: dimension to access output (for ASR, TTS, MT oaxis=0, reserved for future research, -1 means all axis.) :return: List[List[Tuple[str, dict]]] list of batches z arg 'count' (z) should be one of zarg 'batch_sort_key' (rrrrr]r^r_r`z*cannot detect `count` manually set one of zcount is auto detected as raz;batch_sort_key=shuffle is only available if batch_count=seqcategorycst|dddS)Nrrrr)r[batch_sort_axisbatch_sort_keyr r r szmake_batchset..)keyrr5) rr r!r"r#r$r%r&r')rCr"r#r$r&)rPrQrRr"r#r$r&NrX)BATCH_COUNT_CHOICESrBATCH_SORT_KEY_CHOICESr@r+rZ setdefaultgetvaluesr\rsortedrArr4rOrWlist itertoolschain)r[rr r!rDr"r#rfswap_iomtcountrCbatch_frames_inbatch_frames_outbatch_frames_inoutr%r'r$r& category2datakv batches_listdbatchesrr rdr make_batchsetGs>           r})rFrrrr)rrFrr) ror@numpyrr4rOrWr\rhrifloatr}r r r r sT T ` m#