https://github.com/sannetenoever/STiMCON
Tip revision: 873a2bf5c79fe2f828e72e14ef74db409d387854 authored by Sanne ten Oever on 23 July 2021, 08:31:05 UTC
Update README.md
Update README.md
Tip revision: 873a2bf
RNN_subFun.cpython-37.pyc
B
bÒÕ`&: ã @ sÞ d Z ddlZddlZddlZddlmZ ddlmZm Z m
Z
m Z m Z m
Z
ddlZddlmZ ddlmZ ddlmZ dd d
Zd d Zd
d Zd ddZg dg fddZd ddZG dd dejjZ G dd dejjZ dS ) z6
Created on Mon Mar 16 10:41:49 2020
@author: sanoev
é N)Ú
Sequential)ÚDenseÚDropoutÚ
ActivationÚLSTMÚ
BidirectionalÚ Embedding)Úpadded_everygram_pipeline)ÚMLE)Ú
Vocabularyé c C sÖ t d g }g }g }x>tj t| ¡D ]*}| |¡ | | | ¡ | || ¡ q&W tt| d|d }|d |
||d
}} |d |
||d
}
} t dt| t dt| ||
f| | f||ffS )NzShuffling sentencesg ð?g Y@zSize of training set = %dzSize of test set = %d)ÚprintÚnpÚrandomÚ permutationÚlenÚappendÚint) Zsentences_originalZ
next_originalZpercentage_testZ
tmp_sentencesZtmp_sentences_inxZ
tmp_next_wordÚiZ cut_indexÚx_trainÚx_testÚy_trainÚy_test© r ú/data/workspaces/lacns/workspaces/lacns-sanoev/working_data/Experiments/E002_DNN_N400/01_CompModel/01_Scripts/Git/RNN_subFun.pyÚ shuffle_and_split_training_set s
r c c s d}xt j||ft jd}t j|t jd}xbt|D ]V}x0t| |t| D ]\} }
||
||| f< qNW |||t| ||< |d }q4W ||fV qW d S )Nr )Údtypeé )r ÚzerosÚint32ÚrangeÚ enumerater ) Ú
sentence_listÚnext_word_listÚ
batch_sizeÚ SEQUENCE_LENÚ word_indicesÚindexÚxÚyr ÚtÚwr r r Ú generator) s r, c C s d}t j||ft jd}t j|t jd}xbt|D ]V}x0t| |t| D ]\} }
||
||| f< qLW |||t| ||< |d }q2W ||fS )Nr )r r )r r r r r! r ) r" r# r$ r% r& r' r( r) r r* r+ r r r Úgenerator_return6 s r- c sV d} fddt | D }dd |D }dd |D }t|}t|}| ||¡ |S )Nr c s g | ]\}}| | g qS r r )Ú.0ÚitÚsen)Ú
next_wordsr r ú
<listcomp>D s z(cal_perplexity_model.<locals>.<listcomp>c S s g | ]}t j|d d dqS )F)Ú pad_rightÚpad_left)ÚnltkÚbigrams)r. r* r r r r2 E s c S s g | ]}|D ]}|q qS r r )r. ÚsentÚwordr r r r2 F s )r! r r
Úfit) Ú sentencesr1 ZNgramÚnÚtokenized_textZ
train_dataÚwordsZ padded_vocabZ NLTKmodelr )r1 r Úcal_perplexity_modelA s r> gÉ?c C sº t d t }t|dkr2| tt| dd¡ n<| tt| t|d d¡ |jd |g¡ d|jd _| tt d¡ |dkr| t
|¡ | t t| ¡ | t d¡ |S )NzBuild model...r i, )Ú input_dimÚ
output_dimFé Úsoftmax)
r
r r Úaddr ÚlayersÚ set_weightsÚ trainabler r r r r )r= r% ÚdropoutZEmbLayÚmodelr r r Ú get_modelL s rI ç ð?c C sP t | ¡ d¡} t | ¡| } t | ¡}|t |¡ } t j d| d¡}t |¡S )NÚfloat64r ) r ÚasarrayÚastypeÚlogÚexpÚsumr Ú multinomialÚargmax)ÚpredsÚ temperatureZ exp_predsZprobasr r r Úsample^ s
rU c @ s e Zd Zdd Zdd ZdS )ÚCustomModelEvalc C s@ || _ || _|| _| | _|| _|| _|| _|| _|| _|
| _ d S )N)
r: r1 Úsentences_testÚnext_words_testÚ
examples_fileÚ
meanProb_filer% r& Ú indices_wordÚbigram) ÚselfrY rZ r% r& r[ r: r1 rW rX Ú bigramModelr r r Ú__init__j s zCustomModelEval.__init__c *
sâ j d| ¡ tj tjj ¡}jj | }xødD ]ð}|}j dt| d ¡ j dd |¡ d ¡ j d |¡¡ xt
dD ]}t d j f¡}x&t
|D ]\} }
j|
|d
| f< q²W jj|d
d d
} t| |} j| }
|d d
}| |
¡ j d|
¡ qW j d¡ q>W j d ¡ j ¡ d
}ttj| }t g ¡}t g ¡ xôt
dD ]è}|| ||d g}|| }|dkr¸tjd |d < tj|d
|d
j|d
|d
tj|d
|d
d
j\}}j |¡} t t | d ¡¡ x8t
| jd
D ]&}t || d d
||| f ¡}q8W q|W fddt
jD }dd |D } fddt
jD }dd |D }d
}d
} d
} d
} d
} xt
tjD ]r}||| }j || ¡} j || ¡}!t | ¡dkr,| | } | d } t |!¡dkrÜ| |! } | d } qÜW |tj }"| | }#| | }$tjj tjd
j\}}j |¡} t | d ¡ fddt
jD }dd |D } fddt
jD }dd |D }d
}d
} d
} d
}%d
}&xt
tjD ]z}|| ||| f }j || ¡} j || ¡}!t | ¡dkrt| | } |%d }%t |!¡dkr | |! } |&d }&q W |tj }'| |% }(| |& })j d| ¡ j dt|" d ttj d ¡ j d t|' d ttj d ¡ j d t|# d t| tj d ¡ j d t|( d t|%tj d ¡ j d t|$ d t| tj d ¡ j d!t|) d t|&tj d ¡ j ¡ d S )"Nz'
----- Generating text after Epoch: %d
)g333333Ó?gÙ?g à?g333333ã?gffffffæ?z----- Diversity:Ú
z ----- Generating with seed:
"ú z"
é2 r r )ÚverbosezQ================================================================================
é
é é c s$ g | ] \}}|j | g qS r )r[ )r. r/ r0 )Ú
indMaxWordr] r r r2 © s z0CustomModelEval.on_epoch_end.<locals>.<listcomp>c S s g | ]}t j|d d dqS )F)r3 r4 )r5 r6 )r. r* r r r r2 ª s c s* g | ]"\}}|d gj | g qS )éÿÿÿÿ)r[ )r. r/ r0 )rg r] r r r2 « s c S s g | ]}t j|d d dqS )F)r3 r4 )r5 r6 )r. r* r r r r2 ¬ s Fc s$ g | ] \}}|j | g qS r )r[ )r. r/ r0 )rg r] r r r2 Ä s c S s g | ]}t j|d d dqS )F)r3 r4 )r5 r6 )r. r* r r r r2 Å s c s* g | ]"\}}|d gj | g qS )rh )r[ )r. r/ r0 )rg r] r r r2 Æ s c S s g | ]}t j|d d dqS )F)r3 r4 )r5 r6 )r. r* r r r r2 Ç s z'
----- Prop/Per/Bigram after Epoch: %d
z PropTest ú zPropVal zPerplexicityTest zPerplexityVal z BigramTest z
BigramVal )!rY Úwriter r Úrandintr r: rW ÚstrÚjoinr r r% r! r& rH ÚpredictrU r[ r Úflushr Úarrayr- r1 rR Úshaper\ Ú
perplexityÚmathÚisinfrX rZ )*r] ÚepochÚlogsÚ
seed_indexÚseedÚ diversityÚsentencer Úx_predr* r8 rS Ú
next_indexÚ next_wordÚbsZstepTZAPZbvZrangeVrC Ú
modelInputÚlabelsr/ r< Ú test_dataÚtokenized_text_biÚtest_data_bigramÚadprÚadperÚadbiÚcntperÚcntbiÚperÚbiÚmeanPÚmeanPerÚmeanBiÚ cntperValÚcntbiValÚmeanPValÚ
meanPerValÚ meanBiValr )rg r] r Ú on_epoch_endv s´
P , **....z CustomModelEval.on_epoch_endN)Ú__name__Ú
__module__Ú __qualname__r_ r r r r r rV i s rV c @ s e Zd Zdd Zdd ZdS )ÚCustomModelEval_v2c C s@ || _ || _|| _| | _|| _|| _|| _|| _|| _|
| _ d S )N)
r: r1 rW rX rY rZ r% r& r[ r\ ) r] rY rZ r% r& r[ r: r1 rW rX r^ r r r r_ ä s zCustomModelEval_v2.__init__c $ sÎ j d| ¡ tj tjj ¡}jj | }xødD ]ð}|}j dt| d ¡ j dd |¡ d ¡ j d |¡¡ xt
dD ]}t d j f¡}x&t
|D ]\} }
j|
|d
| f< q²W jj|d
d d
} t| |} j| }
|d d
}| |
¡ j d|
¡ qW j d¡ q>W j d ¡ j ¡ tjjtjd
j\}}j |¡} t | d ¡ fddt
jD }dd |D } fddt
jD }dd |D }d
}d
}d
}d
}d
}xt
tjD ]z}|| ||| f }j || ¡}j || ¡}t |¡dkrL|| }|d }t |¡dkrô|| }|d }qôW |tj } || } || } d
} d
} d
}!d
}"d
}#j d| ¡ j dt| d ttj d ¡ j dt| d ttj d ¡ j dt|! d t| tj d ¡ j dt| d t|tj d ¡ j dt|" d t|#tj d ¡ j dt| d t|tj d ¡ j ¡ d S ) Nz'
----- Generating text after Epoch: %d
)g333333Ó?gÙ?g à?g333333ã?gffffffæ?z----- Diversity:r` z ----- Generating with seed:
"ra z"
rb r r )rc zQ================================================================================
rd c s$ g | ] \}}|j | g qS r )r[ )r. r/ r0 )rg r] r r r2 s z3CustomModelEval_v2.on_epoch_end.<locals>.<listcomp>c S s g | ]}t j|d d dqS )F)r3 r4 )r5 r6 )r. r* r r r r2 s c s* g | ]"\}}|d gj | g qS )rh )r[ )r. r/ r0 )rg r] r r r2 s c S s g | ]}t j|d d dqS )F)r3 r4 )r5 r6 )r. r* r r r r2 s Fz'
----- Prop/Per/Bigram after Epoch: %d
z PropTest ri zPropVal zPerplexicityTest zPerplexityVal z BigramTest z
BigramVal ) rY rj r r rk r r: rW rl rm r r r% r! r& rH rn rU r[ r ro r- rX rR r\ rr rs rt rZ r1 )$r] ru rv rw rx ry rz r r{ r* r8 rS r| r} r r r< r r r r r
r r r r/ r r r r r r r r r r r )rg r] r r ð sv
**....z CustomModelEval_v2.on_epoch_endN)r r r r_ r r r r r r ã s r )r )r )rJ ) Ú__doc__rs Únumpyr ÚkerasZ keras.modelsr Z keras.layersr r r r r r r5 Znltk.lm.preprocessingr Znltk.lmr
r r r, r- r> rI rU Ú callbacksÚCallbackrV r r r r r Ú<module> s"
z