Slightly cleaner version of joining with spaces. Args: tokens (list<string>): the sentence to print Return: st

Extract the sample lines from the conversations Args: conversation (Obj): a conversation object containing the lines to extract

Load the pre-processed full corpus and filter the vocabulary / sentences to match the given model options

Prepare the batches for the current epoch Return: list<Batch>: Get a list of the batches for the next epoch

Parse the argument to decide on which device run the model Return: str: The name of the device on which run the program

Convenience method for some tf arguments

Instantiate the chatbot for later use Should be called only once

Args: dirName (str): folder to load Return: array(question, answer): the extracted QA pairs

Args: fileName (str): file to load field (set<str>): fields to extract Return: dict<dict<str>>: t

Load/create the conversations data

Initialize embeddings with pre-trained word2vec vectors Will modify the embedding weights of the current loaded model Uses the Google

Args: fileName (str): file to load Return: list<dict<str>>: the extracted fields for each line

Args: fileName (str): file to load field (set<str>): fields to extract Return: dict<dict<str>>: t

Args: fileName (str): file to load Return: list<dict<str>>: the extracted fields for each line

Args: fileName (str): file to load

Load the some values associated with the current model, like the current globStep value For now, this function does not need to be called bef

Try predicting the sentences that the user will enter in the console Args: sess: The current running session

Training loop Args: sess: The current running session

Only keep a small fraction of the dataset, given by the ratio

Restore or reset the model, depending of the parameters If the destination directory already contains some file, it will handle the conflict

Parse the arguments from the given command line Args: args (list<str>): List of arguments to parse. If None, the default

Print a random dialogue from the dataset

Try predicting the sentences from the samples.txt file. The sentences are saved on the modelDir under the same name Args:

Save the params of the model, like the current globStep value Warning: if you modify this function, make sure the changes mirror loadModelPar

Encode a sequence and return a batch as an input for the model Return: Batch: a batch object containing the sentence, or none if s

Shuffle the training samples

Forward an iterable eventually wrapped around a tqdm decorator The iterable is only wrapped if the iterable contains enough elements Args:

Project the output of the decoder into the vocabulary space Args: X (tf.Tensor): input value

Load all conversations Args: args: parameters of the model

Args: shape: a tuple (input dim, output dim) scope (str): encapsulate variables dtype: the weights type

Args: args: parameters of the model textData: the dataset object

Args: dirName (string): directory where to load the corpus

Args: dirName (string): directory where to load the corpus

Args: dirName (string): directory where to load the corpus

Args: dirName (string): directory where to load the corpus

Args: lightweightFile (string): file containing our lightweight-formatted corpus

A utility function to close the daemon when finish

Generator over the mini-batch training samples

Main view which launch and handle the chatbot view Args: request (Obj): django request object

Merge the sentences until the max sentence length is reached Also decrement id count for unused sentences. Args:

Print a complete batch, useful for debugging Args: batch (Batch): a batch object

Called when a client try to open a WebSocket Args: message (Obj): object containing the client query

Called when a client disconnect Args: message (Obj): object containing the client query

Called when a client send a message Args: message (Obj): object containing the client query