Prediction using LSTM

This module contains the code necessary to make a recommendation based on a LSTM.

Functions

get_latest_file_with_path(path, *paths):

Method to get the full path name for the latest file for the input parameter in paths. This method uses the os.path.getctime function to get the most recently created file that matches the filename pattern in the provided path.

Parameters:

pathstring

Root pathname for the files.

*pathsstring list

These are the var args field, the optional set of strings to denote the full path to the file names.

Returns:

latest_filestring

Full path name for the latest file provided in the paths parameter.

Class

LSTMpredict

Functions

predict(self, device, net, words, vocab_to_int, int_to_vocab, names_set):

Returns the list of top 5 predictions for the provided list of words using the model stored in net. The device is initialized to CPU for the purpose of the predictions.

Parameters:

devicetorch.device

Device type to signify ‘cpu’ or ‘gpu’.

nettorch.module

Instance of LSTM created using RNN Module.

wordslist

List of strings used for predicting the next string in the list of words.

vocab_to_intdict

Mapping of strings to int used to embed the input strings.

int_to_vocabdict

Mapping of int to string used in the process of running the model predictions.

names_setdict

Mapping of fieldType(example proxyObsType, interpretation/variable) and list of all the possible values the field can take.

Returns:

list

Top 5 recommendations for the next string in the sequence of words.

predictForSentence(self, sentence, isInferred = False):

This method is used from the Flask Server Code API. This method handles the initialization of the lstm model for the two different prediction chains we are using in our system.

archiveType -> proxyObservationType -> units archiveType -> proxyObservationType -> interpretation/variable -> interpretation/variableDetial -> inferredVariable -> inferredVarUnits

Depending on the length of the input sentence and the variableType, it chooses the output that will be returned to the server.

If the variableType == measured then we will be considering the complete chain for prediction example: If sentence length = 1, it contains the archiveType and output = prediction for proxyObservationType example: If sentence length = 2, it contains the archiveType and proxyObservationType and output = units and interpretation/variable so on..

If the variableType == inferred then we will be considering the top attributes in the chain from proxyObservationType to interpretation/variableDetail to predict the top 5 inferredVariable as the output .

Parameters:

sentencestring

Input sentence to predict the next field.

isInferredboolean, optional

True if variableType == ‘inferred’. The default is False.

Returns:

dict

Contains the result list of predictions as the value. Depending on the length of the input sentence and the variableType, the dict can contain one item corresponding to key ‘0’ or two items corresponding to the two keys ‘0’ and ‘1’.

Usage

LSTMpredict.py module is used for accuracy calculation in the /accuracy_calc/lstm directory. For more information check out the documentation for Metrics Calculation.

• Evaluation Metrics for LSTM