the-algorithm/trust_and_safety_models/nsfw/nsfw_text.py

161 lines
4.1 KiB
Python
Raw Normal View History

2023-04-17 06:19:03 +02:00
import os
import re
from datetime import datetime
from functools import reduce
2023-04-17 06:19:03 +02:00
import matplotlib.pyplot as plt
import pandas as pd
import tensorflow as tf
2023-04-17 06:19:03 +02:00
from sklearn.metrics import (
PrecisionRecallDisplay,
average_precision_score,
classification_report,
precision_recall_curve,
)
from sklearn.model_selection import train_test_split
from twitter.cuad.representation.models.optimization import create_optimizer
from twitter.cuad.representation.models.text_encoder import TextEncoder
2023-04-17 06:19:03 +02:00
pd.set_option("display.max_colwidth", None)
pd.set_option("display.expand_frame_repr", False)
print(tf.__version__)
print(tf.config.list_physical_devices())
2023-04-17 06:19:03 +02:00
log_path = os.path.join(
"pnsfwtweettext_model_runs", datetime.now().strftime("%Y-%m-%d_%H.%M.%S")
)
2023-04-17 06:19:03 +02:00
tweet_text_feature = "text"
params = {
2023-04-17 06:19:03 +02:00
"batch_size": 32,
"max_seq_lengths": 256,
"model_type": "twitter_bert_base_en_uncased_augmented_mlm",
"trainable_text_encoder": True,
"lr": 5e-5,
"epochs": 10,
}
REGEX_PATTERNS = [
2023-04-17 06:19:03 +02:00
r"^RT @[A-Za-z0-9_]+: ",
r"@[A-Za-z0-9_]+",
2023-04-17 06:19:03 +02:00
r"https:\/\/t\.co\/[A-Za-z0-9]{10}",
r"@\?\?\?\?\?",
]
EMOJI_PATTERN = re.compile(
"(["
"\U0001F1E0-\U0001F1FF"
"\U0001F300-\U0001F5FF"
"\U0001F600-\U0001F64F"
"\U0001F680-\U0001F6FF"
"\U0001F700-\U0001F77F"
"\U0001F780-\U0001F7FF"
"\U0001F800-\U0001F8FF"
"\U0001F900-\U0001F9FF"
"\U0001FA00-\U0001FA6F"
"\U0001FA70-\U0001FAFF"
"\U00002702-\U000027B0"
"])"
2023-04-17 06:19:03 +02:00
)
def clean_tweet(text):
for pattern in REGEX_PATTERNS:
2023-04-17 06:19:03 +02:00
text = re.sub(pattern, "", text)
text = re.sub(EMOJI_PATTERN, r" \1 ", text)
text = re.sub(r"\n", " ", text)
return text.strip().lower()
2023-04-17 06:19:03 +02:00
df["processed_text"] = df["text"].astype(str).map(clean_tweet)
df.sample(10)
2023-04-17 06:19:03 +02:00
X_train, X_val, y_train, y_val = train_test_split(
df[["processed_text"]], df["is_nsfw"], test_size=0.1, random_state=1
)
def df_to_ds(X, y, shuffle=False):
2023-04-17 06:19:03 +02:00
ds = tf.data.Dataset.from_tensor_slices(
(X.values, tf.one_hot(tf.cast(y.values, tf.int32), depth=2, axis=-1))
)
if shuffle:
ds = ds.shuffle(1000, seed=1, reshuffle_each_iteration=True)
return ds.map(lambda text, label: ({tweet_text_feature: text}, label)).batch(
params["batch_size"]
)
ds_train = df_to_ds(X_train, y_train, shuffle=True)
ds_val = df_to_ds(X_val, y_val)
X_train.values
inputs = tf.keras.layers.Input(shape=(), dtype=tf.string, name=tweet_text_feature)
encoder = TextEncoder(
2023-04-17 06:19:03 +02:00
max_seq_lengths=params["max_seq_lengths"],
model_type=params["model_type"],
trainable=params["trainable_text_encoder"],
local_preprocessor_path="demo-preprocessor",
)
embedding = encoder([inputs])["pooled_output"]
2023-04-17 06:19:03 +02:00
predictions = tf.keras.layers.Dense(2, activation="softmax")(embedding)
model = tf.keras.models.Model(inputs=inputs, outputs=predictions)
model.summary()
optimizer = create_optimizer(
2023-04-17 06:19:03 +02:00
params["lr"],
params["epochs"] * len(ds_train),
0,
weight_decay_rate=0.01,
optimizer_type="adamw",
)
bce = tf.keras.losses.BinaryCrossentropy(from_logits=False)
2023-04-17 06:19:03 +02:00
pr_auc = tf.keras.metrics.AUC(curve="PR", num_thresholds=1000, from_logits=False)
model.compile(optimizer=optimizer, loss=bce, metrics=[pr_auc])
callbacks = [
2023-04-17 06:19:03 +02:00
tf.keras.callbacks.EarlyStopping(
monitor="val_loss", mode="min", patience=1, restore_best_weights=True
),
tf.keras.callbacks.ModelCheckpoint(
filepath=os.path.join(log_path, "checkpoints", "{epoch:02d}"), save_freq="epoch"
),
tf.keras.callbacks.TensorBoard(
log_dir=os.path.join(log_path, "scalars"),
update_freq="batch",
write_graph=False,
),
]
history = model.fit(
2023-04-17 06:19:03 +02:00
ds_train,
epochs=params["epochs"],
callbacks=callbacks,
validation_data=ds_val,
steps_per_epoch=len(ds_train),
)
model.predict(["xxx 🍑"])
preds = X_val.processed_text.apply(apply_model)
print(classification_report(y_val, preds >= 0.90, digits=4))
precision, recall, thresholds = precision_recall_curve(y_val, preds)
fig = plt.figure(figsize=(15, 10))
plt.plot(precision, recall, lw=2)
plt.grid()
plt.xlim(0.2, 1)
plt.ylim(0.3, 1)
plt.xlabel("Recall", size=20)
plt.ylabel("Precision", size=20)
average_precision_score(y_val, preds)