AI

Summurizing 结构 请为文章提供一个结构化的概要 理解 请介绍一下该技术的背景与发展历史。 请介绍一下主要特性与关键技术。 请列举上述介绍中涉及到的技术概念，请详细阐述涉及的每个概念，并举例说明。 （请重点解释上述概念间的差异与对比。） 使用 请介绍如何安装相关依赖或工具。 请介绍具体使用方式，并给出相应例子。 （简单介绍一下实际应用案例。） （使用中遇到哪些常见问题，如何解决。） 拓展 有没有类似工具，请对比其优劣 格式 请检查文章的格式，要求语句流畅，无错别字，格式美观，中英文间有空格，变量或命令用反括号。请直接返回校正后的文章。 Code Reading README 主要内容 依赖包，以及它们的功能 项目结构 文档、wiki Learning Summarising Summarise the above text into the most important points. Display the points as bullet points with short descriptions. Mind Map Create a mind map on the topic above. List out the central idea, main branches, and sub-branches. Active Recall I’ve recently immersed myself in studying [specific topic]. To ensure I’ve truly grasped the essence and intricacies, could you challenge my understanding by posing some deep, thought-provoking questions? This will not only test my recall of the main concepts and principles but also solidify my grasp on the finer details. ...

PyTorch PyTorch - QuickStart Basic # version torch.__version__ # PyTorch version torch.version.cuda # Corresponding CUDA version torch.backends.cudnn.version() # Corresponding cuDNN version torch.cuda.get_device_name(0) # GPU type # seed torch.manual_seed(0) torch.cuda.manual_seed_all(0) # GPU torch.cuda.is_available() os.environ['CUDA_VISIBLE_DEVICES'] = '0,1' torch.cuda.empty_cache() # clear GPU cache device = ( "cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu" ) Tensor tensor = torch.randn(2, 3) # tensor info tensor.type() # Data type tensor.size() # Shape of the tensor tensor.shape # Shape of the tensor tensor.dim() # Number of dimensions # type convertions tensor = tensor.cuda() tensor = tensor.cpu() tensor = tensor.float() tensor = tensor.long() torch.set_default_tensor_type(torch.FloatTensor) # Set default tensor type # torch.Tensor <=> np.ndarray ndarray = tensor.cpu().numpy() tensor = torch.from_numpy(ndarray).float() tensor = torch.from_numpy(ndarray.copy()).float() # If ndarray has negative stride # 从只包含一个元素的张量中提取值 value = tensor.item() # torch.Tensor <=> PIL.Image # PyTorch中的张量默认采用N×D×H×W的顺序，并且数据范围在[0, 1]，需要进行转置和规范化。 image = PIL.Image.fromarray(torch.clamp(tensor * 255, min=0, max=255 ).byte().permute(1, 2, 0).cpu().numpy()) image = torchvision.transforms.functional.to_pil_image(tensor) # Equivalently way tensor = torch.from_numpy(np.asarray(PIL.Image.open(path)) ).permute(2, 0, 1).float() / 255 tensor = torchvision.transforms.functional.to_tensor(PIL.Image.open(path)) # Equivalently way # np.ndarray <=> PIL.Image image = PIL.Image.fromarray(ndarray.astypde(np.uint8)) ndarray = np.asarray(PIL.Image.open(path)) # reshape tensor.reshape(shape) tensor.permute(0,2,1) # swap axes tensor.flatten() # 展成 1D 向量 tensor.squeeze() # 去掉维数为 1 的的维度 tensor.unsqueeze(dim=0) # 增加维度 # shuffle tensor = tensor[torch.randperm(tensor.size(0))] # Shuffle the first dimension # copy # Operation | New/Shared memory | Still in computation graph | tensor.clone() # | New | Yes | tensor.detach() # | Shared | No | tensor.detach().clone() # | New | No | # concatenate tensor = torch.cat(list_of_tensors, dim=0) tensor = torch.stack(list_of_tensors, dim=0) # one-hot N = tensor.size(0) one_hot = torch.zeros(N, num_classes).long() one_hot.scatter_(dim=1, index=torch.unsqueeze(tensor, dim=1), src=torch.ones(N, num_classes).long()) # non-zero torch.nonzero(tensor) # Index of non-zero elements torch.nonzero(tensor == 0) # Index of zero elements torch.nonzero(tensor).size(0) # Number of non-zero elements torch.nonzero(tensor == 0).size(0) # Number of zero elements # equal torch.allclose(tensor1, tensor2) # float tensor torch.equal(tensor1, tensor2) # int tensor # expand # Expand tensor of shape 64*512 to shape 64*512*7*7. torch.reshape(tensor, (64, 512, 1, 1)).expand(64, 512, 7, 7) # normalize F.normalize(tensor, dim=1) Dataloader from torch.utils.data import TensorDataset, Dataset, DataLoader train_ds = TensorDataset(x_train[:50000], y_train[:50000]) valid_ds = TensorDataset(x_train[50000:], y_train[50000:]) train_dl = DataLoader(train_ds, batch_size=mini_batch, shuffle=True, num_workers=4) valid_dl = DataLoader(valid_ds, batch_size=len(valid_ds)) for xb, yb in train_dl: pass 自定义dataset ...

References Github - awesome-rl Github - awesome-deep-rl Github - rl-book Unity ML-Agents Toolkit Github - train_your_own_game_AI Gymnasium Github - CleanRL Frameworks Github - CleanRL Github - Stable Baselines3 Autonomous Vehicle Github - CARLA Github - CARLA-SB3-RL-Training-Environment LLM Github - trl Github - Verl Github - ROLL

sklearn-Autograd sklearn Prepocessing from sklearn.preprocessing import LabelEncoder, OneHotEncoder, LabelBinarizer # feature (OneHotEncoder) v.s. labels (LabelBinarizer) data = ['cold', 'cold', 'warm', 'cold', 'hot', 'hot', 'warm', 'cold', 'warm', 'hot'] LabelEncoder().fit_transform(data) # -> array([0, 0, 2, 0, 1, 1, 2, 0, 2, 1]) OneHotEncoder(sparse=False).fit_transform(np.array(data).reshape(-1,1)) # -> array([[1., 0., 0.], # [1., 0., 0.], # ..., # [0., 1., 0.]]) LabelBinarizer().fit_transform(data) # -> array([[1, 0, 0], # [1, 0, 0], # ..., # [0, 1, 0]]) # one label v.s. multilabels data = [["US", "M"], ["UK", "M"], ["FR", "F"]] OneHotEncoder(sparse=False).fit_transform(data) # -> array([[0., 0., 1., 0., 1.], # [0., 1., 0., 0., 1.], # [1., 0., 0., 1., 0.]]) MultiLabelBinarizer().fit_transform(data) # -> array([[0, 0, 1, 0, 1], # [0, 0, 1, 1, 0], # [1, 1, 0, 0, 0]]) Feature Selection sklearn - Feature selection 知乎 - 特征选择方法全面总结 Filter Methods Pearson’s Correlation ...

Basic # define constant a = tf.constant(1) b = tf.constant(2, name='const', shape=(3,5), dtype=tf.float64) # define operation add = a + b mul = tf.square(tf.multiply(a, b)) # define placeholder p = tf.placeholder(tf.int32) # define variables x = tf.get_variable('x', [], dtype=tf.int32) y = tf.get_variable('y', shape=(3,5), initializer=tf.constant_initializer(0)) # session a = tf.constant(1) b = tf.constant(2) f = a + b sess = tf.Session() print(sess.run(f)) # constant do not need initialization x = tf.get_variable('x', [], dtype=tf.int32) f = x + a sess = tf.Session() sess.run(x.initializer) # variable must be initialized before run # return multiple items in a single sess.run() call, instead of making multiple calls print(sess.run([x, f])) p = tf.placeholder(tf.int32) sess = tf.Session() print(sess.run(p, feed_dict={p:2})) # placeholder must be feed in run # assignment # assign node is not connected to the variable, it change the variable's value by side effect x = tf.get_variable('x', [], dtype=tf.int32) a = tf.constant(1) assign = tf.assign(x, a) sess = tf.Session() sess.run(assign) print(sess.run(x)) # sess with block x = tf.get_variable('x', shape=(3,5), initializer=tf.constant_initializer(0)) y = tf.get_variable('y', shape=(3,5), initializer=tf.constant_initializer(0)) f = x + y with tf.Session() as sess: x.initializer.run() y.initializer.run() print(f.eval()) x = tf.get_variable('x', shape=(3,5), initializer=tf.constant_initializer(0)) y = tf.get_variable('y', shape=(3,5), initializer=tf.constant_initializer(0)) f = x * x g = f + y with tf.Session() as sess: x.initializer.run() y.initializer.run() f_val, g_val = sess.run([f, g]) # eval f and g in one run print(f_val, g_val) # interative session x = tf.get_variable('x', [], initializer=tf.constant_initializer(3)) f = x * x sess = tf.InteractiveSession() x.initializer.run() print(f.eval()) # global initializer x = tf.get_variable('x', shape=[], initializer=tf.constant_initializer(1)) y = tf.get_variable('y', shape=[], initializer=tf.constant_initializer(2)) f = x + y init = tf.global_variables_initializer() with tf.Session() as sess: init.run() print(f.eval()) # name scope x = tf.get_variable('x', [], , initializer=tf.constant_initializer(1)) with tf.variable_scope('scope'): y = tf.get_variable('x', [], , initializer=tf.constant_initializer(2)) print(x.name, y.name) # get graph default_graph = tf.get_default_graph() print(default_graph) new_graph = tf.Graph() with new_graph.as_default(): x = tf.Variable(2) print(x.graph) Save and Load a Model # save a model x = tf.get_variable('x', []) y = tf.get_variable('y', []) init = tf.global_variables_initializer() # define the saver after every variable is defined saver = tf.train.Saver() sess = tf.Session() sess.run(init) # save the model after run saver.save(sess, './models/test-model') # load a model x = tf.get_variable('x', []) y = tf.get_variable('y', []) saver = tf.train.Saver() sess = tf.Session() saver.restore(sess, './models/test-model') # no need to init sess.run([x, y]) Optimization # autodiff x = tf.get_variable('x', [], initializer=tf.constant_initializer(3.)) y = tf.get_variable('y', [], initializer=tf.constant_initializer(2.)) f = y*x + x*x + 3*x grad = tf.gradients(f, [x, y]) sess = tf.Session() sess.run(x.initializer) sess.run(y.initializer) print(sess.run(grad)) # return [df/dx, df/dy] # example k = tf.get_variable('k', [], initializer=tf.constant_initializer(0.)) b = tf.get_variable('b', [], initializer=tf.constant_initializer(0.)) init = tf.global_variables_initializer() x = tf.placeholder(tf.float32) y = tf.placeholder(tf.float32) y_pred = k * x + b loss = tf.square(y - y_pred) optimizer = tf.train.GradientDescentOptimizer(1e-3) train_op = optimizer.minimize(loss) sess = tf.Session() sess.run(init) import random true_k = random.random() true_b = random.random() for update_i in range(10000): input_data = random.random() output_data = random.random() _loss, _ = sess.run([loss, train_op], feed_dict={x: input_data, y: output_data}) print(update_i, _loss) print('True parameter: k={}, b={}'.format(true_k, true_b)) print('True parameter: k={}, b={}'.format(sess.run([m, b]))) Tensorboard from datetime import datetime now = datetime.now().strftime("%Y%m%d%H%M%S") root_logdir = "tf_logs" logdir = "{}/run-{}/".format(root_logdir, now) # end of construction phase mse_summary = tf.summary.scalar('MSE', mse) file_writer = tf.summary.FileWriter(logdir, tf.get_default_graph()) # execution phase for batch_index in range(n_batches): X_batch, y_batch = fetch_batch(epoch, batch_index, batch_size) if batch_index % 10 == 0: summary_str = mse_summary.eval(feed_dict={X: X_batch, y: y_batch}) step = epoch * n_batches + batch_index file_writer.add_summary(summary_str, step) sess.run(training_op, feed_dict={X: X_batch, y: y_batch}) # finally file_writer.close() # start the Tensorboard service !tensorboard --logdir tf_logs/ Keras sequential model import keras from keras.models import Sequential from keras.layers import Dense (x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data() y_train = keras.utils.to_categorical(y_train) model = Sequential() model.add(Dense(100, activation='relu', input_shape=(28*28,))) model.add(Dense(100, activation='relu')) model.add(Dense(10, activation='softmax')) model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['acc']) history = model.fit(x_train.reshape(-1, 28*28), y_train, validation_split=0.1, batch_size=32, epochs=10) general model import keras from keras.layers import Input, Dense from keras.models import Model (x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data() y_train = keras.utils.to_categorical(y_train) input_tensor = Input(shape=(28*28, )) x = Dense(100, activation='relu')(input_tensor) x = Dense(100, activation='relu')(x) output_tensor = Dense(10, activation='softmax')(x) model = Model(input_tensor, output_tensor) model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['acc']) history = model.fit(x_train.reshape(-1, 28*28), y_train, validation_split=0.1, batch_size=32, epochs=10) Use Keras Layers in Tensorflow import tensorflow as tf from tensorflow import keras from tensorflow.keras.layers import Dense x = tf.placeholder(name='x', shape=(None, 28*28), dtype=tf.float32) hidden1 = Dense(100, activation='relu')(x) hidden2 = Dense(100, activation='relu')(hidden1) output = Dense(10)(hidden2) y = tf.placeholder(tf.int64, shape=(None,)) xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y, logits=output) loss = tf.reduce_mean(xentropy, name="loss") # train with tensorflow # ... Use Keras Model in Tensorflow import tensorflow as tf from tensorflow import keras from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense model = Sequential() model.add(Dense(100, activation='relu', input_shape=(28*28,))) model.add(Dense(10, activation='softmax')) x = tf.placeholder(name='input', shape=(None, 28*28), dtype=tf.float32) y = model(x) Check GPU # check if tensorflow is gpu version pip list | grep tensorflow from keras import backend as K K.tensorflow_backend._get_available_gpus() from tensorflow.python.client import device_lib print(device_lib.list_local_devices()) import tensorflow as tf sess = tf.Session(config=tf.ConfigProto(log_device_placement=True)) with tf.device('/gpu:0'): a = tf.constant([1., 2., 3.], shape=[1,3], name='a') b = tf.constant([1., 2., 3.], shape=[3,1], name='b') c = tf.matmul(a, b) with tf.Session() as sess: print(sess.run(c))