User:Francis Tyers/Perceptron
A perceptron is a classifier that
The classifier consists of:
- An input vector,
- A weight vector,
- Threshold,
The input vector is made up of binary features, such as:
Training algorithm[edit]
The objective of the training algorithm is to find the most adequate set of weights, and a threshold .
def decision(input, weights, threshold): #{
#}
# Initialise weights and threshold.
weights = [0.0, 0.0];
threshold = 0.0;
errors = 0;
while True: #{
# If there are no errors, training has converged.
if errors == 0: #{
break;
#}
#}
Example[edit]
Here is a worked example of a perceptron applied to the task of lexical selection. Lexical selection is the task of choosing a target translation for a given source word in a context out of a set of possible translations . A perceptron makes a classification decision for a single class, so we need to train a separate perceptron for each possible target word selection.
In the example,
- = estació
- = {season, station}
- = season
Features[edit]
The features we will be working with are ngram contexts around the "problem word". These can be extracted from the word alignments calculated from a parallel corpus.
Catalan | English |
---|---|
Durant l' estació seca les pluges són escasses. | During the dry season it rains infrequently. |
L' estiu és una estació de l' any. | Summer of one of the seasons of the year. |
Barcelona-Sants és una estació de tren a Barcelona. | Barcelona-Sants is a train station in Barcelona. |
... |
Training data[edit]
- s = estació, t* = season
_ sec |
1 |
_ de el any |
1 |
_ de tren |
0 |
_ de el línia |
0 |
_ humit |
1 |
_ plujós |
1 |
un _ a |
0 |
Feature vector[edit]
This is the above training data expressed as an input vector to the perceptron.
_ sec | _ de el any | _ de tren | _ de la línia | _ humit | _ plujós | un _ a | ||
---|---|---|---|---|---|---|---|---|
1 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | |
0 | 1 | 0 | 0 | 0 | 0 | 0 | 1 | |
0 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | |
0 | 0 | 0 | 1 | 0 | 0 | 0 | 0 | |
0 | 0 | 0 | 0 | 1 | 0 | 0 | 1 | |
0 | 0 | 0 | 0 | 0 | 1 | 0 | 1 | |
0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 |
Weight vector[edit]
_ sec | _ de el any | _ de tren | _ de la línia | _ humit | _ plujós | un _ a |
0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |