mirror of
https://github.com/Steffo99/unimore-bda-6.git
synced 2024-11-25 17:24:20 +00:00
91 lines
3.2 KiB
Python
91 lines
3.2 KiB
Python
import tensorflow
|
|
import itertools
|
|
import typing as t
|
|
|
|
from ..database import DataSet, Text, Category
|
|
from ..tokenizer import BaseTokenizer
|
|
from .base import BaseSentimentAnalyzer, AlreadyTrainedError, NotTrainedError
|
|
|
|
|
|
class TensorflowSentimentAnalyzer(BaseSentimentAnalyzer):
|
|
def __init__(self, *, tokenizer: BaseTokenizer):
|
|
super().__init__(tokenizer=tokenizer)
|
|
self.trained = False
|
|
self.text_vectorization_layer = None
|
|
self.neural_network: tensorflow.keras.Sequential | None = None
|
|
|
|
@staticmethod
|
|
def __infinite_dataset_generator_factory(dataset: DataSet):
|
|
"""
|
|
A generator of infinite copies of dataset.
|
|
|
|
.. todo:: Loads the whole dataset in memory. What a waste! Can we perform multiple MongoDB queries instead?
|
|
"""
|
|
dataset = map(lambda text, category: (tensorflow.convert_to_tensor(text, dtype=tensorflow.string), tensorflow.convert_to_tensor(category, dtype=tensorflow.string)), dataset)
|
|
|
|
def generator():
|
|
while True:
|
|
nonlocal dataset
|
|
dataset, result = itertools.tee(dataset, 2)
|
|
yield result
|
|
|
|
return generator
|
|
|
|
@classmethod
|
|
def __bda_dataset_to_tf_dataset(cls, dataset: DataSet) -> tensorflow.data.Dataset:
|
|
"""
|
|
Convert a `unimore_bda_6.database.DataSet` to a "real" `tensorflow.data.Dataset`.
|
|
"""
|
|
return tensorflow.data.Dataset.from_generator(
|
|
cls.__infinite_dataset_generator_factory(dataset),
|
|
output_signature=(
|
|
tensorflow.TensorSpec(shape=(), dtype=tensorflow.string),
|
|
tensorflow.TensorSpec(shape=(), dtype=tensorflow.string),
|
|
)
|
|
)
|
|
|
|
MAX_FEATURES = 20000
|
|
EMBEDDING_DIM = 16
|
|
EPOCHS = 10
|
|
|
|
def train(self, training_set: DataSet) -> None:
|
|
if self.trained:
|
|
raise AlreadyTrainedError()
|
|
|
|
training_set = self.__bda_dataset_to_tf_dataset(training_set)
|
|
|
|
self.text_vectorization_layer = tensorflow.keras.layers.TextVectorization(
|
|
max_tokens=self.MAX_FEATURES,
|
|
standardize=self.tokenizer.tokenize_tensorflow,
|
|
)
|
|
self.text_vectorization_layer.adapt(map(lambda t: t[0], training_set))
|
|
|
|
training_set = training_set.map(self.text_vectorization_layer)
|
|
|
|
# I have no idea of what I'm doing here
|
|
self.neural_network = tensorflow.keras.Sequential([
|
|
tensorflow.keras.layers.Embedding(self.MAX_FEATURES + 1, self.EMBEDDING_DIM),
|
|
tensorflow.keras.layers.Dropout(0.2),
|
|
tensorflow.keras.layers.GlobalAveragePooling1D(),
|
|
tensorflow.keras.layers.Dropout(0.2),
|
|
tensorflow.keras.layers.Dense(1),
|
|
])
|
|
|
|
self.neural_network.compile(
|
|
loss=tensorflow.losses.BinaryCrossentropy(from_logits=True), # Only works with two tags
|
|
metrics=tensorflow.metrics.BinaryAccuracy(threshold=0.0)
|
|
)
|
|
|
|
self.neural_network.fit(
|
|
training_set,
|
|
epochs=self.EPOCHS,
|
|
)
|
|
|
|
self.trained = True
|
|
|
|
def use(self, text: Text) -> Category:
|
|
if not self.trained:
|
|
raise NotTrainedError()
|
|
|
|
prediction = self.neural_network.predict(text)
|
|
breakpoint()
|