nmt_utils.cpython-36.pyc

3

��BZ�@s�dd
lZ
ddlmZ
dd
lZddlmZ
ddlmZ
ddlm	Z	
dd
l
jZe�Z
e
jd�

ejd�

ddd	d
ddd
ddddddddgZdg
Zdd�Zdd�Zdd�Zdd�Zddd d!gZd"d#�Zef
d$d%�
Zd*d&d'�
Zd+d(d)�
Zd
S),�N)
�Faker)
�tqdm)
�format_date)
�to_categoricali90�short�medium�long�fullz	d MMM YYYz
d MMMM YYYz
dd MMM YYYz
d MMM, YYYzd MMMM, YYYzdd, MMM YYYzd MM YYz
MMMM d YYYzMMMM d, YYYzdd.MM.YY�en_UScCs�tj
�}yXt|tjt�
tjt�
d
�}
tjdddg�
}|dkrF|
j�}
n|dkrV|
j�}
|j	�}Wn t
k
r�
}
zdSd}~XnX|
||fS)z�
        Creates some fake dates 
        :returns: tuple containing human readable string, machine readable string, and date object
    )�format�localer
�
�N)NNN)�fake�date_objectr�random�choice�FORMATS�LOCALES�upper�lower�	isoformat�AttributeError)�dtZhuman_readableZcase_changeZmachine_readable�
e�r�@/home/jovyan/work/Week 3/Neural Machine Translation/nmt_utils.py�create_date"s







rc
s"
t�}
t�}g}xRt
t|�
�
D]B}t�\}}}|d
k	r|j||f�

|
jt|�
�

|jt|�
�

qWttt	|
�
ddgt	tt
|
�
d�
�
��
�
ttt	|�
ddg�
�
}dd�|j�D�
�d�t|�\}	}
t
j��
fdd	�|	D�
�
}	��fd
d	�|
D�
}
t
jt	t�f
dd�|
��
�
}
|�
�||	|
fS)
zx
        Creates a dataset with n_examples and vocabularies
        :n_examples: the number of examples to generate
    Nz<unk>z<pad>rc
Ssi|]\}
}|
|�qSrr)�.0�
k�
vrrr�
<dictcomp>Nsz"create_dataset.<locals>.<dictcomp>�c
sg|]}
t|
��
��qSr)
�
string_to_int)r�
i)�Tx�humanrr�
<listcomp>Tsz"create_dataset.<locals>.<listcomp>c
sg|]}
t|
��
��qSr)
r#)r�
t)r%�machinerrr'Usc

st|t
��
d
�S)N)
�num_classes)r�len)
�
x)
r)rr�<lambda>Vsz create_dataset.<locals>.<lambda>)�setr�ranger�append�update�tuple�dict�zip�listr+�	enumerate�items�np�array�map)Z
n_examplesZhuman_vocabZ
machine_vocab�datasetr$�
h�
m�
_Zinv_machine�sources�targetsr)r%r&r)r�create_dataset:s&











rAcsXt|�
|
kr|d
|
�}t
t�f
dd�|��
}t|�
|
krT|�dg
|
t|�
7}|S)a$
    Converts all strings in the vocabulary into a list of integers representing the positions of the
    input string's characters in the "vocab"
    
    Arguments:
    string -- input string, e.g. 'Wed 10 Jul 2007'
    length -- the number of time steps you'd like, determines if the output will be padded or cut
    vocab -- vocabulary, dictionary used to index every character of your "string"
    
    Returns:
    rep -- list of integers (or '<unk>') (size = length) representing the position of the string's character in the vocabulary
    Nc

s�j|d
�S)Nz<unk>)
�get)
r,)
�vocabrrr-mszstring_to_int.<locals>.<lambda>z<pad>)r+r5r:)�string�lengthrC�repr)
rCrr#\s

r#cs�f
d
d�|D�
}|S)a�

    Output a machine readable list of characters based on a list of indexes in the machine's vocabulary
    
    Arguments:
    ints -- list of integers representing indexes in the machine's vocabulary
    inv_vocab -- dictionary mapping machine readable indexes to machine readable characters 
    
    Returns:
    l -- list of characters corresponding to the indexes of ints thanks to the inv_vocab mapping
    c
sg|]}
�|
�qSrr)rr$)
�	inv_vocabrrr'�sz!int_to_string.<locals>.<listcomp>r)�intsrG�
lr)
rGr�
int_to_stringus
rJz
3 May 1979z5 Apr 09z20th February 2016zWed 10 Jul 2007cCs:t|t
|
�}|jtj|g
�
�
}tj|d
dd�}t||�S)Nr
r
)
�axis�����)r#r%�predictr8r9�argmaxrJ)�model�input_vocabulary�inv_output_vocabulary�text�encoded�
predictionrrr�run_example�s



rUcCsHg}x>|D]6}|jd
j
t||
||��
�

td|�
td|d�
q
W|S)N�zinput:zoutput:r
rL)r0�joinrU�print)rOrPrQ�examplesZ	predicted�examplerrr�run_examples�s







r[cCst|d
krtj
jd|d�}tj
j||
|fd�
}tj
jdd|dfd�}tj|d
d
�d|f�|d
d
�|d
d
�f<||fS)aS
    Data generation. x is purely random except that it's first value equals the target y.
    In practice, the network should learn that the target = x[attention_column].
    Therefore, most of its attention should be focused on the value addressed by attention_column.
    :param m: the number of samples to retrieve.
    :param Tx: the number of time steps of your series.
    :param n_h: the number of dimensions of each element in the series.
    :param attention_column: the column linked to the target. Everything else is purely random.
    :return: x: model inputs, y: model targets
    Nr
)�low�high)
�sizerr
)r\r]r^)r8r�randint�standard_normal�tile)r=r%Zn_hZattention_columnr,�
yrrr�get_data_recurrent�s



*rccsxg}|j��dkr$d
d�|j
D�
}n�f
dd�|j
D�
}�f
dd�|D�
}�
f
dd�|D�
}x|D]}|j|�

qbW|S)Nc
Ssg|]
}
|
j�qSr)
�output)r�layerrrrr'�sz#get_activations.<locals>.<listcomp>c
sg|]}
|
j�kr|
j
�qSr)�namerd)rre)
�
layer_namerrr'�sc
s&g|]}
tj
�g
tj�g
|
g
��qSr)�
K�function�learning_phase)r�out)
�inprrr'�sc
sg|]}
|
�dg�
d
�qS)g�?r
r)r�func)
�inputsrrr'�s)�input�layersr0)rOrnrg�activations�outputs�funcsZ
layer_outputsZlayer_activationsr)rlrnrgr�get_activations�s









rt)
N)
N)�numpyr8�fakerrrr�babel.datesr�keras.utilsr�
keras.backend�backendrhr�seedrrrrAr#rJZEXAMPLESrUr[rcrtrrrr�<module>
sD
























"