from keras.callbacks import Callback<\/p>\nclass EarlyStoppingByLossVal(Callback):<\/p>\n
def __init__(self, monitor='val_loss', value=0.0001, verbose=0):<\/p>\n
super(Callback, self).__init__()<\/p>\n
self.monitor = monitor<\/p>\n
self.value = value<\/p>\n
self.verbose = verbose<\/p>\n
def on_epoch_end(self, epoch, logs={}):<\/p>\n
current = logs.get(self.monitor)<\/p>\n
if current is None:<\/p>\n
warnings.warn("Early stopping requires %s available!" % self.monitor, RuntimeWarning)<\/p>\n
if current < self.value:<\/p>\n
if self.verbose > 0:<\/p>\n
print(f"Epoch {epoch}: early stopping")<\/p>\n
self.model.stop_training = True<\/p>\n
early_stopping = EarlyStoppingByLossVal(monitor='val_loss', value=0.0001, verbose=1)<\/p>\n
model.fit(x_train, y_train, callbacks=[early_stopping])<\/p>\n
<\/code><\/pre>\n<\/p>\n
2.2 \u4f7f\u7528\u5185\u7f6e\u56de\u8c03<\/h4>\n<\/p>\n
\u8bb8\u591a\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u63d0\u4f9b\u4e86\u5185\u7f6e\u7684\u56de\u8c03\u51fd\u6570\uff0c\u5982Keras\u7684EarlyStopping\uff0c\u53ef\u4ee5\u65b9\u4fbf\u5730\u5b9e\u73b0\u65e9\u505c\u3002<\/p>\n<\/p>\n
from keras.callbacks import EarlyStopping<\/p>\nearly_stopping = EarlyStopping(monitor='val_loss', patience=2)<\/p>\n
model.fit(x_train, y_train, callbacks=[early_stopping])<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u4e09\u3001\u4fdd\u5b58\u8bad\u7ec3\u72b6\u6001<\/h3>\n<\/p>\n
\u5728\u957f\u65f6\u95f4\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u4e3a\u4e86\u9632\u6b62\u610f\u5916\u4e2d\u65ad\u5bfc\u81f4\u7684\u8bad\u7ec3\u8fdb\u5ea6\u4e22\u5931\uff0c\u53ef\u4ee5\u4fdd\u5b58\u8bad\u7ec3\u72b6\u6001\u3002\u8fd9\u79cd\u65b9\u6cd5\u4e0d\u4ec5\u53ef\u4ee5\u7528\u4e8e\u6682\u505c\u8bad\u7ec3\uff0c\u8fd8\u53ef\u4ee5\u5728\u9700\u8981\u65f6\u6062\u590d\u8bad\u7ec3\u3002<\/p>\n<\/p>\n
3.1 \u4fdd\u5b58\u68c0\u67e5\u70b9<\/h4>\n<\/p>\n
\u901a\u8fc7\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5b9a\u671f\u4fdd\u5b58\u6a21\u578b\u68c0\u67e5\u70b9\uff0c\u53ef\u4ee5\u5728\u8bad\u7ec3\u4e2d\u65ad\u540e\u4ece\u6700\u8fd1\u7684\u68c0\u67e5\u70b9\u6062\u590d\u8bad\u7ec3\u3002\u8fd9\u79cd\u65b9\u6cd5\u901a\u5e38\u4e0e\u56de\u8c03\u51fd\u6570\u7ed3\u5408\u4f7f\u7528\u3002<\/p>\n<\/p>\n
from keras.callbacks import ModelCheckpoint<\/p>\ncheckpoint = ModelCheckpoint(filepath='model.h5', save_best_only=True)<\/p>\n
model.fit(x_train, y_train, callbacks=[checkpoint])<\/p>\n
<\/code><\/pre>\n<\/p>\n
3.2 \u6062\u590d\u8bad\u7ec3<\/h4>\n<\/p>\n
\u5728\u8bad\u7ec3\u4e2d\u65ad\u540e\uff0c\u53ef\u4ee5\u52a0\u8f7d\u4fdd\u5b58\u7684\u6a21\u578b\u68c0\u67e5\u70b9\uff0c\u7ee7\u7eed\u8bad\u7ec3\u3002<\/p>\n<\/p>\n
from keras.models import load_model<\/p>\nmodel = load_model('model.h5')<\/p>\n
model.fit(x_train, y_train)<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u56db\u3001\u624b\u52a8\u4e2d\u65ad<\/h3>\n<\/p>\n
\u5728\u67d0\u4e9b\u60c5\u51b5\u4e0b\uff0c\u7528\u6237\u53ef\u80fd\u5e0c\u671b\u624b\u52a8\u4e2d\u65ad\u8bad\u7ec3\u4ee5\u8fdb\u884c\u8c03\u8bd5\u6216\u5176\u4ed6\u64cd\u4f5c\u3002\u8fd9\u53ef\u4ee5\u901a\u8fc7\u4fe1\u53f7\u5904\u7406\u6216\u624b\u52a8\u8bbe\u7f6e\u6807\u5fd7\u6765\u5b9e\u73b0\u3002<\/p>\n<\/p>\n
4.1 \u4fe1\u53f7\u5904\u7406<\/h4>\n<\/p>\n
\u53ef\u4ee5\u901a\u8fc7\u6355\u83b7\u4fe1\u53f7\uff08\u5982SIGINT\uff09\u6765\u5b9e\u73b0\u5b89\u5168\u4e2d\u65ad\u3002<\/p>\n<\/p>\n
import signal<\/p>\ndef signal_handler(signal, frame):<\/p>\n
print('Training interrupted')<\/p>\n
global interrupted<\/p>\n
interrupted = True<\/p>\n
signal.signal(signal.SIGINT, signal_handler)<\/p>\n
interrupted = False<\/p>\n
for epoch in range(epochs):<\/p>\n
if interrupted:<\/p>\n
break<\/p>\n
# Training code here<\/p>\n
<\/code><\/pre>\n<\/p>\n
4.2 \u624b\u52a8\u6807\u5fd7<\/h4>\n<\/p>\n
\u901a\u8fc7\u8bbe\u7f6e\u5168\u5c40\u53d8\u91cf\u6216\u4f7f\u7528\u5171\u4eab\u5185\u5b58\uff0c\u53ef\u4ee5\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u52a8\u6001\u8c03\u6574\u8bad\u7ec3\u72b6\u6001\u3002<\/p>\n<\/p>\n
stop_training = False<\/p>\nfor epoch in range(epochs):<\/p>\n
if stop_training:<\/p>\n
break<\/p>\n
# Training code here<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u4e94\u3001\u52a8\u6001\u8c03\u6574\u53c2\u6570<\/h3>\n<\/p>\n
\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u52a8\u6001\u8c03\u6574\u8bad\u7ec3\u53c2\u6570\uff08\u5982\u5b66\u4e60\u7387\u3001\u6279\u91cf\u5927\u5c0f\uff09\u4e5f\u53ef\u4ee5\u5e2e\u52a9\u63a7\u5236\u8bad\u7ec3\u8fdb\u5ea6\uff0c\u5e76\u5728\u5fc5\u8981\u65f6\u6682\u505c\u8bad\u7ec3\u3002<\/p>\n<\/p>\n
5.1 \u52a8\u6001\u5b66\u4e60\u7387<\/h4>\n<\/p>\n
\u901a\u8fc7\u5b66\u4e60\u7387\u8c03\u5ea6\uff0c\u53ef\u4ee5\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u6839\u636e\u6761\u4ef6\u52a8\u6001\u8c03\u6574\u5b66\u4e60\u7387\uff0c\u8fdb\u800c\u63a7\u5236\u8bad\u7ec3\u901f\u5ea6\u3002<\/p>\n<\/p>\n
from keras.callbacks import LearningRateScheduler<\/p>\ndef scheduler(epoch, lr):<\/p>\n
if epoch < 10:<\/p>\n
return lr<\/p>\n
else:<\/p>\n
return lr * 0.1<\/p>\n
lr_scheduler = LearningRateScheduler(scheduler)<\/p>\n
model.fit(x_train, y_train, callbacks=[lr_scheduler])<\/p>\n
<\/code><\/pre>\n<\/p>\n
5.2 \u52a8\u6001\u6279\u91cf\u5927\u5c0f<\/h4>\n<\/p>\n
\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u52a8\u6001\u8c03\u6574\u6279\u91cf\u5927\u5c0f\u4e5f\u53ef\u4ee5\u5e2e\u52a9\u4f18\u5316\u8bad\u7ec3\u6548\u7387\u3002<\/p>\n<\/p>\n
# Example: Adjust batch size dynamically<\/p>\ninitial_batch_size = 32<\/p>\n
for epoch in range(epochs):<\/p>\n
if condition: # Define your condition to change batch size<\/p>\n
batch_size = new_batch_size<\/p>\n
else:<\/p>\n
batch_size = initial_batch_size<\/p>\n
# Training code here with dynamic batch size<\/p>\n
<\/code><\/pre>\n<\/p>\n
\u516d\u3001\u603b\u7ed3<\/h3>\n<\/p>\n
\u6682\u505c\u6a21\u578b\u8bad\u7ec3\u5728\u6df1\u5ea6\u5b66\u4e60\u4e2d\u662f\u4e00\u4e2a\u91cd\u8981\u7684\u6280\u80fd\uff0c\u901a\u8fc7\u8bbe\u7f6e\u8bad\u7ec3\u8f6e\u6570\u3001\u4f7f\u7528\u56de\u8c03\u51fd\u6570\u3001\u4fdd\u5b58\u8bad\u7ec3\u72b6\u6001\u3001\u624b\u52a8\u4e2d\u65ad\u4ee5\u53ca\u52a8\u6001\u8c03\u6574\u53c2\u6570\uff0c\u53ef\u4ee5\u6709\u6548\u5730\u63a7\u5236\u548c\u4f18\u5316\u8bad\u7ec3\u8fc7\u7a0b\u3002\u6bcf\u79cd\u65b9\u6cd5\u5404\u6709\u5176\u9002\u7528\u573a\u666f\u548c\u4f18\u52bf\uff0c\u7528\u6237\u53ef\u4ee5\u6839\u636e\u5177\u4f53\u9700\u6c42\u9009\u62e9\u5408\u9002\u7684\u65b9\u6cd5\u6765\u5b9e\u73b0\u8bad\u7ec3\u7684\u6682\u505c\u4e0e\u6062\u590d\u3002\u603b\u7684\u6765\u8bf4\uff0c\u7075\u6d3b\u8fd0\u7528\u8fd9\u4e9b\u6280\u5de7\uff0c\u53ef\u4ee5\u5927\u5e45\u63d0\u5347\u6a21\u578b\u8bad\u7ec3\u7684\u6548\u7387\u548c\u6548\u679c\u3002<\/p>\n<\/p>\n
\u76f8\u5173\u95ee\u7b54FAQs\uff1a<\/strong><\/h2>\n \u5982\u4f55\u5728Python\u4e2d\u6682\u505c\u6a21\u578b\u8bad\u7ec3\u7684\u8fc7\u7a0b\uff1f<\/strong>
\u5728Python\u4e2d\uff0c\u53ef\u4ee5\u901a\u8fc7\u8bbe\u7f6e\u6807\u5fd7\u6216\u4f7f\u7528\u56de\u8c03\u51fd\u6570\u6765\u5b9e\u73b0\u6a21\u578b\u8bad\u7ec3\u7684\u6682\u505c\u3002\u5e38\u89c1\u7684\u65b9\u6cd5\u662f\u4f7f\u7528\u4e00\u4e2a\u5e03\u5c14\u53d8\u91cf\u6765\u63a7\u5236\u8bad\u7ec3\u5faa\u73af\u7684\u6267\u884c\uff0c\u6216\u8005\u5728\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u4e2d\u5229\u7528ModelCheckpoint<\/code>\u6216EarlyStopping<\/code>\u7b49\u56de\u8c03\u51fd\u6570\u3002\u8fd9\u4e9b\u65b9\u6cd5\u53ef\u4ee5\u5e2e\u52a9\u4f60\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u968f\u65f6\u6682\u505c\u6a21\u578b\u3002<\/p>\n\u6682\u505c\u8bad\u7ec3\u540e\u5982\u4f55\u6062\u590d\u6a21\u578b\u8bad\u7ec3\uff1f<\/strong>
\u6062\u590d\u8bad\u7ec3\u901a\u5e38\u6d89\u53ca\u5230\u4fdd\u5b58\u548c\u52a0\u8f7d\u6a21\u578b\u7684\u72b6\u6001\u3002\u5728\u5927\u591a\u6570\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u4e2d\uff08\u5982TensorFlow\u6216PyTorch\uff09\uff0c\u53ef\u4ee5\u5728\u6682\u505c\u65f6\u4fdd\u5b58\u5f53\u524d\u6a21\u578b\u7684\u6743\u91cd\u548c\u4f18\u5316\u5668\u7684\u72b6\u6001\uff0c\u4e4b\u540e\u901a\u8fc7\u52a0\u8f7d\u8fd9\u4e9b\u72b6\u6001\u6765\u7ee7\u7eed\u8bad\u7ec3\u3002\u786e\u4fdd\u5728\u6062\u590d\u8bad\u7ec3\u65f6\u4fdd\u6301\u76f8\u540c\u7684\u5b66\u4e60\u7387\u548c\u5176\u4ed6\u8d85\u53c2\u6570\u8bbe\u7f6e\u3002<\/p>\n\u662f\u5426\u53ef\u4ee5\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u52a8\u6001\u8c03\u6574\u6a21\u578b\u7684\u53c2\u6570\uff1f<\/strong>
\u662f\u7684\uff0c\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u53ef\u4ee5\u52a8\u6001\u8c03\u6574\u6a21\u578b\u7684\u53c2\u6570\u3002\u8fd9\u53ef\u4ee5\u901a\u8fc7\u8bbe\u7f6e\u56de\u8c03\u51fd\u6570\u6765\u5b9e\u73b0\uff0c\u4f8b\u5982\u5728Keras\u4e2d\u4f7f\u7528ReduceLROnPlateau<\/code>\u6765\u52a8\u6001\u8c03\u6574\u5b66\u4e60\u7387\u3002\u6b64\u5916\uff0c\u8fd8\u53ef\u4ee5\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u901a\u8fc7\u76d1\u63a7\u635f\u5931\u51fd\u6570\u6216\u5176\u4ed6\u6307\u6807\u6765\u51b3\u5b9a\u4f55\u65f6\u8c03\u6574\u6a21\u578b\u7684\u8d85\u53c2\u6570\uff0c\u5982\u6279\u91cf\u5927\u5c0f\u6216\u5c42\u6570\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"\u5728Python\u4e2d\uff0c\u8bad\u7ec3\u6a21\u578b\u65f6\u53ef\u4ee5\u901a\u8fc7\u8bbe\u7f6e\u8bad\u7ec3\u8f6e\u6570\u3001\u6dfb\u52a0\u56de\u8c03\u51fd\u6570\u3001\u4fdd\u5b58\u8bad\u7ec3\u72b6\u6001\u3001\u624b\u52a8\u4e2d\u65ad\u7b49\u65b9\u5f0f\u6682\u505c\u6a21\u578b\u8bad\u7ec3\u3002\u8fd9\u4e9b […]","protected":false},"author":3,"featured_media":1002116,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[37],"tags":[],"acf":[],"_links":{"self":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/1002107"}],"collection":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/comments?post=1002107"}],"version-history":[{"count":"1","href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/1002107\/revisions"}],"predecessor-version":[{"id":1002118,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/1002107\/revisions\/1002118"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media\/1002116"}],"wp:attachment":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media?parent=1002107"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/categories?post=1002107"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/tags?post=1002107"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![\"python\u8bad\u7ec3\u6a21\u578b\u5982\u4f55\u6682\u505c\"](\"https:\/\/cdn-kb.worktile.com\/kb\/wp-content\/uploads\/2024\/04\/25080004\/42340789-0e19-48a4-9b71-b5c44a666dd1.webp\")
<\/p>\n