pytorch

探索pytorch

dir()

dir(torch)

['AVG', 'AggregationType', 'AliasDb', 'Any', 'AnyType', 'Argument', 'ArgumentSpec', 'AwaitType', 'BFloat16Storage', 'BFloat16Tensor', 'BenchmarkConfig', 'BenchmarkExecutionStats', 'Block', 'BoolStorage', 'BoolTensor', 'BoolType', 'BufferDict', 'ByteStorage', 'ByteTensor', 'CallStack', 'Callable', 'Capsule', 'CharStorage', 'CharTensor', 'ClassType', 'Code', 'CompilationUnit', 'CompleteArgumentSpec', 'ComplexDoubleStorage', 'ComplexFloatStorage', 'ComplexType', 'ConcreteModuleType', 'ConcreteModuleTypeBuilder', 'DeepCopyMemoTable', 'DeserializationStorageContext', 'DeviceObjType', 'Dict', 'DictType', 'DisableTorchFunction', 'DisableTorchFunctionSubclass', 'DispatchKey', 'DispatchKeySet', 'DoubleStorage', 'DoubleTensor', 'EnumType', 'ErrorReport', 'ExcludeDispatchKeyGuard', 'ExecutionPlan', 'FatalError', 'FileCheck', 'FloatStorage', 'FloatTensor', 'FloatType', 'FunctionSchema', 'Future', 'FutureType', 'Generator', 'Gradient', 'Graph', 'GraphExecutorState', 'HalfStorage', 'HalfTensor', 'IODescriptor', 'InferredType', 'IntStorage', 'IntTensor', 'IntType', 'InterfaceType', 'JITException', 'ListType', 'LiteScriptModule', 'LockingLogger', 'LoggerBase', 'LongStorage', 'LongTensor', 'ModuleDict', 'Node', 'NoneType', 'NoopLogger', 'NumberType', 'OperatorInfo', 'Optional', 'OptionalType', 'PRIVATE_OPS', 'ParameterDict', 'PyObjectType', 'PyTorchFileReader', 'PyTorchFileWriter', 'QInt32Storage', 'QInt8Storage', 'QUInt2x4Storage', 'QUInt4x2Storage', 'QUInt8Storage', 'RRefType', 'SUM', 'ScriptClass', 'ScriptClassFunction', 'ScriptDict', 'ScriptDictIterator', 'ScriptDictKeyIterator', 'ScriptFunction', 'ScriptList', 'ScriptListIterator', 'ScriptMethod', 'ScriptModule', 'ScriptModuleSerializer', 'ScriptObject', 'ScriptObjectProperty', 'SerializationStorageContext', 'Set', 'ShortStorage', 'ShortTensor', 'Size', 'StaticModule', 'Storage', 'StorageBase', 'Stream', 'StreamObjType', 'StringType', 'SymBool', 'SymFloat', 'SymInt', 'SymIntType', 'TYPE_CHECKING', 'Tag', 'Tensor', 'TensorType', 'ThroughputBenchmark', 'TracingState', 'TupleType', 'Type', 'TypedStorage', 'USE_GLOBAL_DEPS', 'USE_RTLD_GLOBAL_WITH_LIBTORCH', 'Union', 'UnionType', 'UntypedStorage', 'Use', 'Value', '_C', '_GLOBAL_DEVICE_CONTEXT', '_TorchCompileInductorWrapper', '_VF', '__all__', '__annotations__', '__builtins__', '__cached__', '__config__', '__doc__', '__file__', '__future__', '__loader__', '__name__', '__package__', '__path__', '__spec__', '__version__', '_adaptive_avg_pool2d', '_adaptive_avg_pool3d', '_add_batch_dim', '_add_relu', '_add_relu_', '_addmm_activation', '_aminmax', '_amp_foreach_non_finite_check_and_unscale_', '_amp_update_scale_', '_assert', '_assert_async', '_assert_tensor_metadata', '_awaits', '_batch_norm_impl_index', '_cast_Byte', '_cast_Char', '_cast_Double', '_cast_Float', '_cast_Half', '_cast_Int', '_cast_Long', '_cast_Short', '_choose_qparams_per_tensor', '_chunk_grad_outputs_efficient_attention', '_classes', '_coalesce', '_compute_linear_combination', '_conj', '_conj_copy', '_conj_physical', '_convert_indices_from_coo_to_csr', '_convert_indices_from_csr_to_coo', '_convolution', '_convolution_mode', '_copy_from', '_copy_from_and_resize', '_ctc_loss', '_cudnn_ctc_loss', '_cudnn_init_dropout_state', '_cudnn_rnn', '_cudnn_rnn_flatten_weight', '_cufft_clear_plan_cache', '_cufft_get_plan_cache_max_size', '_cufft_get_plan_cache_size', '_cufft_set_plan_cache_max_size', '_cummax_helper', '_cummin_helper', '_debug_has_internal_overlap', '_decomp', '_dim_arange', '_dirichlet_grad', '_disable_functionalization', '_dispatch', '_efficientzerotensor', '_embedding_bag', '_embedding_bag_forward_only', '_empty_affine_quantized', '_empty_per_channel_affine_quantized', '_enable_functionalization', '_euclidean_dist', '_fake_quantize_learnable_per_channel_affine', '_fake_quantize_learnable_per_tensor_affine', '_fake_quantize_per_tensor_affine_cachemask_tensor_qparams', '_fft_c2c', '_fft_c2r', '_fft_r2c', '_foobar', '_foreach_abs', '_foreach_abs_', '_foreach_acos', '_foreach_acos_', '_foreach_add', '_foreach_add_', '_foreach_addcdiv', '_foreach_addcdiv_', '_foreach_addcmul', '_foreach_addcmul_', '_foreach_asin', '_foreach_asin_', '_foreach_atan', '_foreach_atan_', '_foreach_ceil', '_foreach_ceil_', '_foreach_clamp_max', '_foreach_clamp_max_', '_foreach_clamp_min', '_foreach_clamp_min_', '_foreach_cos', '_foreach_cos_', '_foreach_cosh', '_foreach_cosh_', '_foreach_div', '_foreach_div_', '_foreach_erf', '_foreach_erf_', '_foreach_erfc', '_foreach_erfc_', '_foreach_exp', '_foreach_exp_', '_foreach_expm1', '_foreach_expm1_', '_foreach_floor', '_foreach_floor_', '_foreach_frac', '_foreach_frac_', '_foreach_lerp', '_foreach_lerp_', '_foreach_lgamma', '_foreach_lgamma_', '_foreach_log', '_foreach_log10', '_foreach_log10_', '_foreach_log1p', '_foreach_log1p_', '_foreach_log2', '_foreach_log2_', '_foreach_log_', '_foreach_maximum', '_foreach_maximum_', '_foreach_minimum', '_foreach_minimum_', '_foreach_mul', '_foreach_mul_', '_foreach_neg', '_foreach_neg_', '_foreach_norm', '_foreach_reciprocal', '_foreach_reciprocal_', '_foreach_round', '_foreach_round_', '_foreach_sigmoid', '_foreach_sigmoid_', '_foreach_sin', '_foreach_sin_', '_foreach_sinh', '_foreach_sinh_', '_foreach_sqrt', '_foreach_sqrt_', '_foreach_sub', '_foreach_sub_', '_foreach_tan', '_foreach_tan_', '_foreach_tanh', '_foreach_tanh_', '_foreach_trunc', '_foreach_trunc_', '_foreach_zero_', '_freeze_functional_tensor', '_from_functional_tensor', '_functorch', '_fused_adam_', '_fused_adamw_', '_fused_dropout', '_fused_moving_avg_obs_fq_helper', '_fused_sdp_choice', '_fw_primal_copy', '_grid_sampler_2d_cpu_fallback', '_guards', '_has_compatible_shallow_copy_type', '_histogramdd_bin_edges', '_histogramdd_from_bin_cts', '_histogramdd_from_bin_tensors', '_import_dotted_name', '_index_put_impl_', '_indices_copy', '_initExtension', '_is_all_true', '_is_any_true', '_is_functional_tensor', '_is_zerotensor', '_jit_internal', '_linalg_check_errors', '_linalg_det', '_linalg_eigh', '_linalg_slogdet', '_linalg_solve_ex', '_linalg_svd', '_linalg_utils', '_load_global_deps', '_lobpcg', '_log_softmax', '_log_softmax_backward_data', '_logcumsumexp', '_lowrank', '_lstm_mps', '_lu_with_info', '_make_dual', '_make_dual_copy', '_make_per_channel_quantized_tensor', '_make_per_tensor_quantized_tensor', '_masked_scale', '_masked_softmax', '_meta_registrations', '_mkldnn', '_mkldnn_reshape', '_mkldnn_transpose', '_mkldnn_transpose_', '_mps_convolution', '_mps_convolution_transpose', '_namedtensor_internals', '_native_batch_norm_legit', '_native_decoder_only_multi_head_attention', '_native_multi_head_attention', '_neg_view', '_neg_view_copy', '_nested_from_padded', '_nested_from_padded_and_nested_example', '_nested_tensor_from_mask', '_nested_tensor_from_mask_left_aligned', '_nested_tensor_from_tensor_list', '_nested_tensor_softmax_with_shape', '_nnpack_available', '_nnpack_spatial_convolution', '_ops', '_pack_padded_sequence', '_pad_packed_sequence', '_pin_memory', '_preload_cuda_deps', '_prelu_kernel', '_prims', '_prims_common', '_refs', '_register_device_module', '_remove_batch_dim', '_reshape_alias_copy', '_reshape_from_tensor', '_resize_output_', '_rowwise_prune', '_sample_dirichlet', '_saturate_weight_to_fp16', '_scaled_dot_product_attention_math', '_scaled_dot_product_efficient_attention', '_scaled_dot_product_flash_attention', '_segment_reduce', '_shape_as_tensor', '_sobol_engine_draw', '_sobol_engine_ff_', '_sobol_engine_initialize_state_', '_sobol_engine_scramble_', '_softmax', '_softmax_backward_data', '_sources', '_sparse_broadcast_to', '_sparse_broadcast_to_copy', '_sparse_coo_tensor_unsafe', '_sparse_csr_prod', '_sparse_csr_sum', '_sparse_log_softmax_backward_data', '_sparse_softmax_backward_data', '_sparse_sparse_matmul', '_sparse_sum', '_stack', '_standard_gamma', '_standard_gamma_grad', '_storage_classes', '_subclasses', '_sync', '_tensor', '_tensor_classes', '_tensor_str', '_test_autograd_multiple_dispatch', '_test_autograd_multiple_dispatch_view', '_test_autograd_multiple_dispatch_view_copy', '_test_check_tensor', '_test_serialization_subcmul', '_to_cpu', '_to_functional_tensor', '_transform_bias_rescale_qkv', '_transformer_decoder_only_layer_fwd', '_transformer_encoder_layer_fwd', '_trilinear', '_triton_multi_head_attention', '_triton_scaled_dot_attention', '_unique', '_unique2', '_unpack_dual', '_use_cudnn_ctc_loss', '_use_cudnn_rnn_flatten_weight', '_utils', '_utils_internal', '_validate_compressed_sparse_indices', '_validate_sparse_bsc_tensor_args', '_validate_sparse_bsr_tensor_args', '_validate_sparse_compressed_tensor_args', '_validate_sparse_coo_tensor_args', '_validate_sparse_csc_tensor_args', '_validate_sparse_csr_tensor_args', '_values_copy', '_vmap_internals', '_warn_typed_storage_removal', '_weight_norm', '_weight_norm_interface', '_weights_only_unpickler', 'abs', 'abs_', 'absolute', 'acos', 'acos_', 'acosh', 'acosh_', 'adaptive_avg_pool1d', 'adaptive_max_pool1d', 'add', 'addbmm', 'addcdiv', 'addcmul', 'addmm', 'addmv', 'addmv_', 'addr', 'adjoint', 'affine_grid_generator', 'alias_copy', 'align_tensors', 'all', 'allclose', 'alpha_dropout', 'alpha_dropout_', 'amax', 'amin', 'aminmax', 'amp', 'angle', 'any', 'ao', 'arange', 'arccos', 'arccos_', 'arccosh', 'arccosh_', 'arcsin', 'arcsin_', 'arcsinh', 'arcsinh_', 'arctan', 'arctan2', 'arctan_', 'arctanh', 'arctanh_', 'are_deterministic_algorithms_enabled', 'argmax', 'argmin', 'argsort', 'argwhere', 'as_strided', 'as_strided_', 'as_strided_copy', 'as_strided_scatter', 'as_tensor', 'asarray', 'asin', 'asin_', 'asinh', 'asinh_', 'atan', 'atan2', 'atan_', 'atanh', 'atanh_', 'atleast_1d', 'atleast_2d', 'atleast_3d', 'attr', 'autocast', 'autocast_decrement_nesting', 'autocast_increment_nesting', 'autograd', 'avg_pool1d', 'backends', 'baddbmm', 'bartlett_window', 'base_py_dll_path', 'batch_norm', 'batch_norm_backward_elemt', 'batch_norm_backward_reduce', 'batch_norm_elemt', 'batch_norm_gather_stats', 'batch_norm_gather_stats_with_counts', 'batch_norm_stats', 'batch_norm_update_stats', 'bernoulli', 'bfloat16', 'bilinear', 'binary_cross_entropy_with_logits', 'bincount', 'binomial', 'bitwise_and', 'bitwise_left_shift', 'bitwise_not', 'bitwise_or', 'bitwise_right_shift', 'bitwise_xor', 'blackman_window', 'block_diag', 'bmm', 'bool', 'broadcast_shapes', 'broadcast_tensors', 'broadcast_to', 'bucketize', 'builtins', 'can_cast', 'candidate', 'cartesian_prod', 'cat', 'ccol_indices_copy', 'cdist', 'cdouble', 'ceil', 'ceil_', 'celu', 'celu_', 'cfloat', 'chain_matmul', 'chalf', 'channel_shuffle', 'channels_last', 'channels_last_3d', 'cholesky', 'cholesky_inverse', 'cholesky_solve', 'choose_qparams_optimized', 'chunk', 'clamp', 'clamp_', 'clamp_max', 'clamp_max_', 'clamp_min', 'clamp_min_', 'classes', 'classproperty', 'clear_autocast_cache', 'clip', 'clip_', 'clone', 'col_indices_copy', 'column_stack', 'combinations', 'compile', 'compiled_with_cxx11_abi', 'complex', 'complex128', 'complex32', 'complex64', 'concat', 'concatenate', 'conj', 'conj_physical', 'conj_physical_', 'constant_pad_nd', 'contiguous_format', 'conv1d', 'conv2d', 'conv3d', 'conv_tbc', 'conv_transpose1d', 'conv_transpose2d', 'conv_transpose3d', 'convolution', 'copysign', 'corrcoef', 'cos', 'cos_', 'cosh', 'cosh_', 'cosine_embedding_loss', 'cosine_similarity', 'count_nonzero', 'cov', 'cpp', 'cpu', 'cross', 'crow_indices_copy', 'ctc_loss', 'ctypes', 'cuda', 'cuda_path', 'cuda_path_var', 'cuda_version', 'cuda_version_1', 'cudnn_affine_grid_generator', 'cudnn_batch_norm', 'cudnn_convolution', 'cudnn_convolution_add_relu', 'cudnn_convolution_relu', 'cudnn_convolution_transpose', 'cudnn_grid_sampler', 'cudnn_is_acceptable', 'cummax', 'cummin', 'cumprod', 'cumsum', 'cumulative_trapezoid', 'default_generator', 'default_path', 'deg2rad', 'deg2rad_', 'dequantize', 'det', 'detach', 'detach_', 'detach_copy', 'device', 'diag', 'diag_embed', 'diagflat', 'diagonal', 'diagonal_copy', 'diagonal_scatter', 'diff', 'digamma', 'dist', 'distributed', 'distributions', 'div', 'divide', 'dll', 'dll_path', 'dll_paths', 'dlls', 'dot', 'double', 'dropout', 'dropout_', 'dsmm', 'dsplit', 'dstack', 'dtype', 'e', 'eig', 'einsum', 'embedding', 'embedding_bag', 'embedding_renorm_', 'empty', 'empty_like', 'empty_quantized', 'empty_strided', 'enable_grad', 'eq', 'equal', 'erf', 'erf_', 'erfc', 'erfc_', 'erfinv', 'exp', 'exp2', 'exp2_', 'exp_', 'expand_copy', 'expm1', 'expm1_', 'eye', 'fake_quantize_per_channel_affine', 'fake_quantize_per_tensor_affine', 'fbgemm_linear_fp16_weight', 'fbgemm_linear_fp16_weight_fp32_activation', 'fbgemm_linear_int8_weight', 'fbgemm_linear_int8_weight_fp32_activation', 'fbgemm_linear_quantize_weight', 'fbgemm_pack_gemm_matrix_fp16', 'fbgemm_pack_quantized_matrix', 'feature_alpha_dropout', 'feature_alpha_dropout_', 'feature_dropout', 'feature_dropout_', 'fft', 'fill', 'fill_', 'finfo', 'fix', 'fix_', 'flatten', 'flip', 'fliplr', 'flipud', 'float', 'float16', 'float32', 'float64', 'float_power', 'floor', 'floor_', 'floor_divide', 'fmax', 'fmin', 'fmod', 'fork', 'frac', 'frac_', 'frexp', 'frobenius_norm', 'from_dlpack', 'from_file', 'from_numpy', 'frombuffer', 'full', 'full_like', 'func', 'functional', 'fused_moving_avg_obs_fake_quant', 'futures', 'fx', 'gather', 'gcd', 'gcd_', 'ge', 'geqrf', 'ger', 'get_autocast_cpu_dtype', 'get_autocast_gpu_dtype', 'get_default_dtype', 'get_deterministic_debug_mode', 'get_device', 'get_file_path', 'get_float32_matmul_precision', 'get_num_interop_threads', 'get_num_threads', 'get_rng_state', 'glob', 'gradient', 'greater', 'greater_equal', 'grid_sampler', 'grid_sampler_2d', 'grid_sampler_3d', 'group_norm', 'gru', 'gru_cell', 'gt', 'half', 'hamming_window', 'hann_window', 'hardshrink', 'has_cuda', 'has_cudnn', 'has_lapack', 'has_mkl', 'has_mkldnn', 'has_mps', 'has_openmp', 'has_spectral', 'heaviside', 'hinge_embedding_loss', 'histc', 'histogram', 'histogramdd', 'hsmm', 'hsplit', 'hspmm', 'hstack', 'hub', 'hypot', 'i0', 'i0_', 'igamma', 'igammac', 'iinfo', 'imag', 'import_ir_module', 'import_ir_module_from_buffer', 'index_add', 'index_copy', 'index_fill', 'index_put', 'index_put_', 'index_reduce', 'index_select', 'indices_copy', 'inf', 'inference_mode', 'init_num_threads', 'initial_seed', 'inner', 'inspect', 'instance_norm', 'int', 'int16', 'int32', 'int64', 'int8', 'int_repr', 'inverse', 'is_anomaly_check_nan_enabled', 'is_anomaly_enabled', 'is_autocast_cache_enabled', 'is_autocast_cpu_enabled', 'is_autocast_enabled', 'is_complex', 'is_conj', 'is_deterministic_algorithms_warn_only_enabled', 'is_distributed', 'is_floating_point', 'is_grad_enabled', 'is_inference', 'is_inference_mode_enabled', 'is_loaded', 'is_neg', 'is_nonzero', 'is_same_size', 'is_signed', 'is_storage', 'is_tensor', 'is_vulkan_available', 'is_warn_always_enabled', 'isclose', 'isfinite', 'isin', 'isinf', 'isnan', 'isneginf', 'isposinf', 'isreal', 'istft', 'jit', 'kaiser_window', 'kernel32', 'kl_div', 'kron', 'kthvalue', 'last_error', 'layer_norm', 'layout', 'lcm', 'lcm_', 'ldexp', 'ldexp_', 'le', 'legacy_contiguous_format', 'lerp', 'less', 'less_equal', 'lgamma', 'library', 'linalg', 'linspace', 'load', 'lobpcg', 'log', 'log10', 'log10_', 'log1p', 'log1p_', 'log2', 'log2_', 'log_', 'log_softmax', 'logaddexp', 'logaddexp2', 'logcumsumexp', 'logdet', 'logical_and', 'logical_not', 'logical_or', 'logical_xor', 'logit', 'logit_', 'logspace', 'logsumexp', 'long', 'lstm', 'lstm_cell', 'lstsq', 'lt', 'lu', 'lu_solve', 'lu_unpack', 'manual_seed', 'margin_ranking_loss', 'masked', 'masked_fill', 'masked_scatter', 'masked_select', 'math', 'matmul', 'matrix_exp', 'matrix_power', 'matrix_rank', 'max', 'max_pool1d', 'max_pool1d_with_indices', 'max_pool2d', 'max_pool3d', 'maximum', 'mean', 'median', 'memory_format', 'merge_type_from_type_comment', 'meshgrid', 'min', 'minimum', 'miopen_batch_norm', 'miopen_convolution', 'miopen_convolution_add_relu', 'miopen_convolution_relu', 'miopen_convolution_transpose', 'miopen_depthwise_convolution', 'miopen_rnn', 'mkldnn_adaptive_avg_pool2d', 'mkldnn_convolution', 'mkldnn_linear_backward_weights', 'mkldnn_max_pool2d', 'mkldnn_max_pool3d', 'mkldnn_rnn_layer', 'mm', 'mode', 'moveaxis', 'movedim', 'msort', 'mul', 'multinomial', 'multiply', 'multiprocessing', 'mv', 'mvlgamma', 'name', 'nan', 'nan_to_num', 'nan_to_num_', 'nanmean', 'nanmedian', 'nanquantile', 'nansum', 'narrow', 'narrow_copy', 'native_batch_norm', 'native_channel_shuffle', 'native_dropout', 'native_group_norm', 'native_layer_norm', 'native_norm', 'ne', 'neg', 'neg_', 'negative', 'negative_', 'nested', 'nextafter', 'nn', 'no_grad', 'nonzero', 'norm', 'norm_except_dim', 'normal', 'not_equal', 'nuclear_norm', 'numel', 'nvtoolsext_dll_path', 'obj', 'ones', 'ones_like', 'onnx', 'ops', 'optim', 'orgqr', 'ormqr', 'os', 'outer', 'overrides', 'package', 'pairwise_distance', 'parse_ir', 'parse_schema', 'parse_type_comment', 'path_patched', 'pca_lowrank', 'pdist', 'per_channel_affine', 'per_channel_affine_float_qparams', 'per_channel_symmetric', 'per_tensor_affine', 'per_tensor_symmetric', 'permute', 'permute_copy', 'pfiles_path', 'pi', 'pinverse', 'pixel_shuffle', 'pixel_unshuffle', 'platform', 'poisson', 'poisson_nll_loss', 'polar', 'polygamma', 'positive', 'pow', 'prelu', 'prepare_multiprocessing_environment', 'preserve_format', 'prev_error_mode', 'prod', 'profiler', 'promote_types', 'put', 'py_dll_path', 'py_float', 'py_int', 'q_per_channel_axis', 'q_per_channel_scales', 'q_per_channel_zero_points', 'q_scale', 'q_zero_point', 'qint32', 'qint8', 'qr', 'qscheme', 'quantile', 'quantization', 'quantize_per_channel', 'quantize_per_tensor', 'quantize_per_tensor_dynamic', 'quantized_batch_norm', 'quantized_gru', 'quantized_gru_cell', 'quantized_lstm', 'quantized_lstm_cell', 'quantized_max_pool1d', 'quantized_max_pool2d', 'quantized_rnn_relu_cell', 'quantized_rnn_tanh_cell', 'quasirandom', 'quint2x4', 'quint4x2', 'quint8', 'rad2deg', 'rad2deg_', 'rand', 'rand_like', 'randint', 'randint_like', 'randn', 'randn_like', 'random', 'randperm', 'range', 'ravel', 'read_vitals', 'real', 'reciprocal', 'reciprocal_', 'relu', 'relu_', 'remainder', 'renorm', 'repeat_interleave', 'res', 'reshape', 'resize_as_', 'resize_as_sparse_', 'resolve_conj', 'resolve_neg', 'result_type', 'return_types', 'rnn_relu', 'rnn_relu_cell', 'rnn_tanh', 'rnn_tanh_cell', 'roll', 'rot90', 'round', 'round_', 'row_indices_copy', 'row_stack', 'rrelu', 'rrelu_', 'rsqrt', 'rsqrt_', 'rsub', 'saddmm', 'save', 'scalar_tensor', 'scatter', 'scatter_add', 'scatter_reduce', 'searchsorted', 'seed', 'segment_reduce', 'select', 'select_copy', 'select_scatter', 'selu', 'selu_', 'serialization', 'set_anomaly_enabled', 'set_autocast_cache_enabled', 'set_autocast_cpu_dtype', 'set_autocast_cpu_enabled', 'set_autocast_enabled', 'set_autocast_gpu_dtype', 'set_default_device', 'set_default_dtype', 'set_default_tensor_type', 'set_deterministic_debug_mode', 'set_float32_matmul_precision', 'set_flush_denormal', 'set_grad_enabled', 'set_num_interop_threads', 'set_num_threads', 'set_printoptions', 'set_rng_state', 'set_vital', 'set_warn_always', 'sgn', 'short', 'sigmoid', 'sigmoid_', 'sign', 'signal', 'signbit', 'sin', 'sin_', 'sinc', 'sinc_', 'sinh', 'sinh_', 'slice_copy', 'slice_scatter', 'slogdet', 'smm', 'softmax', 'solve', 'sort', 'sparse', 'sparse_bsc', 'sparse_bsc_tensor', 'sparse_bsr', 'sparse_bsr_tensor', 'sparse_compressed_tensor', 'sparse_coo', 'sparse_coo_tensor', 'sparse_csc', 'sparse_csc_tensor', 'sparse_csr', 'sparse_csr_tensor', 'special', 'split', 'split_copy', 'split_with_sizes', 'split_with_sizes_copy', 'spmm', 'sqrt', 'sqrt_', 'square', 'square_', 'squeeze', 'squeeze_copy', 'sspaddmm', 'stack', 'std', 'std_mean', 'stft', 'storage', 'strided', 'sub', 'subtract', 'sum', 'svd', 'svd_lowrank', 'swapaxes', 'swapdims', 'sym_float', 'sym_int', 'sym_max', 'sym_min', 'sym_not', 'symeig', 'sys', 't', 't_copy', 'take', 'take_along_dim', 'tan', 'tan_', 'tanh', 'tanh_', 'tensor', 'tensor_split', 'tensordot', 'testing', 'textwrap', 'th_dll_path', 'threshold', 'threshold_', 'tile', 'to_dlpack', 'topk', 'torch', 'torch_version', 'trace', 'transpose', 'transpose_copy', 'trapezoid', 'trapz', 'triangular_solve', 'tril', 'tril_indices', 'triplet_margin_loss', 'triu', 'triu_indices', 'true_divide', 'trunc', 'trunc_', 'typename', 'types', 'uint8', 'unbind', 'unbind_copy', 'unflatten', 'unfold_copy', 'unify_type_list', 'unique', 'unique_consecutive', 'unsafe_chunk', 'unsafe_split', 'unsafe_split_with_sizes', 'unsqueeze', 'unsqueeze_copy', 'use_deterministic_algorithms', 'utils', 'values_copy', 'vander', 'var', 'var_mean', 'vdot', 'version', 'view_as_complex', 'view_as_complex_copy', 'view_as_real', 'view_as_real_copy', 'view_copy', 'vitals_enabled', 'vmap', 'vsplit', 'vstack', 'wait', 'where', 'windows', 'with_load_library_flags', 'xlogy', 'xlogy_', 'zero_', 'zeros', 'zeros_like']

dir(torch.cuda)

dir(torch.cuda) ['Any', 'BFloat16Storage', 'BFloat16Tensor', 'BoolStorage', 'BoolTensor', 'ByteStorage', 'ByteTensor', 'CUDAGraph', 'CUDAPluggableAllocator', 'CharStorage', 'CharTensor', 'ComplexDoubleStorage', 'ComplexFloatStorage', 'CudaError', 'DeferredCudaCallError', 'Device', 'DoubleStorage', 'DoubleTensor', 'Event', 'ExternalStream', 'FloatStorage', 'FloatTensor', 'HalfStorage', 'HalfTensor', 'IntStorage', 'IntTensor', 'List', 'LongStorage', 'LongTensor', 'Optional', 'OutOfMemoryError', 'ShortStorage', 'ShortTensor', 'Stream', 'StreamContext', 'Tuple', 'Union', '_CudaBase', '_CudaDeviceProperties', '_DeviceGuard', '_LazySeedTracker', '__all__', '__annotations__', '__builtins__', '__cached__', '__doc__', '__file__', '__loader__', '__name__', '__package__', '__path__', '__spec__', '_check_capability', '_check_cubins', '_cudart', '_device', '_device_count_nvml', '_device_t', '_dummy_type', '_exchange_device', '_get_device_index', '_initialization_lock', '_initialized', '_is_compiled', '_is_in_bad_fork', '_lazy_call', '_lazy_init', '_lazy_new', '_lazy_seed_tracker', '_memory_viz', '_nvml_based_avail', '_parse_visible_devices', '_queued_calls', '_raw_device_count_nvml', '_raw_device_uuid_nvml', '_sleep', '_tls', '_transform_uuid_to_ordinals', '_utils', '_warn_typed_storage_removal', 'amp', 'caching_allocator_alloc', 'caching_allocator_delete', 'can_device_access_peer', 'cast', 'change_current_allocator', 'check_error', 'classproperty', 'contextlib', 'cudaStatus', 'cudart', 'current_blas_handle', 'current_device', 'current_stream', 'default_generators', 'default_stream', 'device', 'device_count', 'device_of', 'empty_cache', 'get_allocator_backend', 'get_arch_list', 'get_device_capability', 'get_device_name', 'get_device_properties', 'get_gencode_flags', 'get_rng_state', 'get_rng_state_all', 'get_sync_debug_mode', 'graph', 'graph_pool_handle', 'graphs', 'has_half', 'has_magma', 'init', 'initial_seed', 'ipc_collect', 'is_available', 'is_bf16_supported', 'is_current_stream_capturing', 'is_initialized', 'jiterator', 'list_gpu_processes', 'lru_cache', 'make_graphed_callables', 'manual_seed', 'manual_seed_all', 'max_memory_allocated', 'max_memory_cached', 'max_memory_reserved', 'mem_get_info', 'memory', 'memory_allocated', 'memory_cached', 'memory_reserved', 'memory_snapshot', 'memory_stats', 'memory_stats_as_nested_dict', 'memory_summary', 'memory_usage', 'nccl', 'nvtx', 'os', 'profiler', 'random', 'reset_accumulated_memory_stats', 'reset_max_memory_allocated', 'reset_max_memory_cached', 'reset_peak_memory_stats', 'seed', 'seed_all', 'set_device', 'set_per_process_memory_fraction', 'set_rng_state', 'set_rng_state_all', 'set_stream', 'set_sync_debug_mode', 'sparse', 'stream', 'streams', 'synchronize', 'threading', 'torch', 'traceback', 'utilization', 'warnings']

help()

help(torch.cuda.is_available)

Help on function is_available in module torch.cuda:is_available() -> bool Returns a bool indicating if CUDA is currently available.

可视化数据

dataset提供一种方式获取数据及其label

dataloader

线性回归

权重和偏移量

梯度

学习率

精确解方程:closed form solution

高斯noise 模拟现实

损失函数：（WC+b-y）^2

损失均值

损失极小值点

linear regression

logistic regressionn

加上激活函数（压缩函数）0-1概率问题

classification

正态分布概率密度函数
$$
p(x) = \frac{1}{\sqrt{2 \pi \sigma^2}} \exp\left(-\frac{1}{2 \sigma^2} (x - \mu)^2\right).
$$

tool：

计时器

class Timer:  #@save
    """记录多次运行时间"""
    def __init__(self):
        self.times = []
        self.start()

    def start(self):
        """启动计时器"""
        self.tik = time.time()

    def stop(self):
        """停止计时器并将时间记录在列表中"""
        self.times.append(time.time() - self.tik)
        return self.times[-1]

    def avg(self):
        """返回平均时间"""
        return sum(self.times) / len(self.times)

    def sum(self):
        """返回时间总和"""
        return sum(self.times)

    def cumsum(self):
        """返回累计时间"""
        return np.array(self.times).cumsum().tolist()
        
 
 timer = Timer()
 f'{timer.stop():.5f} sec'

 
 

import torch
x = torch.arange(12)行向量 
x.shape形状 
x.numel()元素的总数
X = x.reshape(3, 4)
X
用x.reshape(-1,4)或x.reshape(3,-1)来取代x.reshape(3,4)。
torch.zeros((2, 3, 4))都是0
torch.ones((2, 3, 4))都是1
torch.randn(3, 4)均值为0、标准差为1的标准高斯分布（正态分布）中随机采样。
torch.tensor([[2, 1, 4, 3], 
              [1, 2, 3, 4], 
              [4, 3, 2, 1]])
X = torch.arange(12, dtype=torch.float32).reshape((3,4))
              
x + y, x - y, x * y, x / y, x ** y
torch.exp(x)
torch.cat((X, Y), dim=0), torch.cat((X, Y), dim=1)沿行（轴-0，形状的第一个元素） 和按列（轴-1，形状的第二个元素）连结两个矩阵
X == Y
X.sum()所有元素进行求和
将两个矩阵广播为一个更大的3×2矩阵，如下所示：矩阵a将复制列， 矩阵b将复制行，然后再按元素相加。


索引和切片
X[-1]选择最后一个元素, X[1:3]选择第二个和第三个元素
指定索引X[1, 2] = 9
[0:2, :]访问第1行和第2行

Z = torch.zeros_like(Y)
print('id(Z):', id(Z))
Z[:] = X + Y
print('id(Z):', id(Z))
用X[:] = X + Y或X += Y来减少操作的内存开销。


A = X.numpy()
B = torch.tensor(A)共享它们的底层内存

a = torch.tensor([3.5])
a, a.item(), float(a), int(a)将大小为1的张量转换为Python标量

数据集
import os

os.makedirs(os.path.join('..', 'data'), exist_ok=True)
data_file = os.path.join('..', 'data', 'house_tiny.csv')
with open(data_file, 'w') as f:
    f.write('NumRooms,Alley,Price\n')  # 列名
    f.write('NA,Pave,127500\n')  # 每行表示一个数据样本
    f.write('2,NA,106000\n')
    f.write('4,NA,178100\n')
    f.write('NA,NA,140000\n')
    
import pandas as pd
data = pd.read_csv(data_file)
print(data)

处理缺失值
inputs, outputs = data.iloc[:, 0:2], data.iloc[:, 2]，先设置输入输出
inputs = inputs.fillna(inputs.mean())填充空为均值

inputs, outputs = data.iloc[:, 0:2], data.iloc[:, 2]
inputs = inputs.fillna(inputs.mean())
inputs = pd.get_dummies(inputs, dummy_na=True)将此列转换为两列“Alley_Pave”和“Alley_nan”

线性代数
x = torch.tensor(3.0)填数字
y = torch.tensor(2.0)
x = torch.arange(4)填长度
A = torch.arange(20).reshape(5, 4)
更多轴
X = torch.arange(24).reshape(2, 3, 4)
tensor([[[ 0,  1,  2,  3],
         [ 4,  5,  6,  7],
         [ 8,  9, 10, 11]],

        [[12, 13, 14, 15],
         [16, 17, 18, 19],
         [20, 21, 22, 23]]])
A = torch.arange(20, dtype=torch.float32).reshape(5, 4)
A.T
B = A.clone()
B == B.T
x + y, x * y, x / y, x**y运算
x[3]索引
len(x)
x.shape
x.sum()
axis降维（轴0）相对抽象，多看
A = np.array([[[1, 2, 3, 4],
               [5, 6, 7, 8],
               [9, 10, 11, 12]],
              
              [[13, 14, 15, 16],
               [17, 18, 19, 20],
               [21, 22, 23, 24]]])
# 沿着第三个维度进行求和（2，3，4）-》（2，3）
result = A.sum(axis=2)降维求和
[[10 26 42]
 [58 74 90]]

A.sum(axis=[0, 1])对行求和不就是剩下列？或者想（2，4，2）-》（2，2）
sum_A = A.sum(axis=1, keepdims=True)非降维求和
    应用 A / sum_A

A.mean(), A.sum() / A.numel()平均值
A.mean(axis=0), A.sum(axis=0) / A.shape[0]

点积
y = torch.ones(4, dtype = torch.float32)
x, y, torch.dot(x, y)
等价torch.sum(x * y)
向量积
A.shape, x.shape, torch.mv(A, x)
矩阵乘法
B = torch.ones(4, 3)
torch.mm(A, B)
范数：
u = torch.tensor([3.0, -4.0])
torch.norm(u)
torch.abs(u).sum()范数1，受异常值的影响较小

$$
范数L2=|\mathbf{x}|2 = \sqrt{\sum{i=1}^n x_i^2},
范数L1=|\mathbf{x}|1 = \sum{i=1}^n \left|x_i \right|.
范数Lp=|\mathbf{x}|p = \left(\sum{i=1}^n \left|x_i \right|^p \right)^{1/p}.
$$

衡量向量的大小或稀疏性，并且在优化问题中经常用于正则化项，以控制模型的复杂度。范数还可以用于计算向量之间的距离，如余弦相似度等。在深度学习中，范数常用于权重衰减（weight decay）和梯度裁剪（gradient clipping）等技术中。

微积分
def f(x):
    return 3 * x ** 2 - 4 * x
def numerical_lim(f, x, h):
    return (f(x + h) - f(x)) / h
    
h = 0.1
for i in range(5):
    print(f'h={h:.5f}, numerical limit={numerical_lim(f, 1, h):.5f}')
    h *= 0.1

绘图
x = np.arange(0, 3, 0.1)
plot(x, [f(x), 2 * x - 3], 'x', 'f(x)', legend=['f(x)', 'Tangent line (x=1)'])


自动微分
x.requires_grad_(True)  # 等价x=torch.arange(4.0,requires_grad=True)
x.grad访问，初始为0
y = 2 * torch.dot(x, x)形成函数也就是 y=2*XT*X,梯度为4*x
y.backward()调用反向传播函数
x.grad输出tensor([ 0.,  4.,  8., 12.])
# 在默认情况下，PyTorch会累积梯度，我们需要清除之前的值
x.grad.zero_()
非标量的反向传播？？
# 对非标量调用backward需要传入一个gradient参数，该参数指定微分函数关于self的梯度。
# 本例只想求偏导数的和，所以传递一个1的梯度是合适的
x.grad.zero_()
y = x * x
# 等价于y.backward(torch.ones(len(x)))
y.sum().backward()
x.grad
梯度不会向后流经u到x
u = y.detach()
z = u * x，u作为常数处理x.grad == u

matplotlib可视化
def use_svg_display():  #@save
    """使用svg格式在Jupyter中显示绘图"""
    backend_inline.set_matplotlib_formats('svg')
    
def set_figsize(figsize=(3.5, 2.5)):  #@save
    """设置matplotlib的图表大小"""
    use_svg_display()
    d2l.plt.rcParams['figure.figsize'] = figsize
#@save
def set_axes(axes, xlabel, ylabel, xlim, ylim, xscale, yscale, legend):轴对象，x标签文本，y标签文本，x列表或元组范围，y，图例的显示否
    """设置matplotlib的轴"""
    axes.set_xlabel(xlabel)
    axes.set_ylabel(ylabel)
    axes.set_xscale(xscale)
    axes.set_yscale(yscale)
    axes.set_xlim(xlim)
    axes.set_ylim(ylim)
    if legend:
        axes.legend(legend)
    axes.grid()
#@save
def plot(X, Y=None, xlabel=None, ylabel=None, legend=None, xlim=None,
         ylim=None, xscale='linear', yscale='linear',
         fmts=('-', 'm--', 'g-.', 'r:'), figsize=(3.5, 2.5), axes=None):
    """绘制数据点"""
    if legend is None:
        legend = []

    set_figsize(figsize)
    axes = axes if axes else d2l.plt.gca()

    # 如果X有一个轴，输出True
    def has_one_axis(X):
        return (hasattr(X, "ndim") and X.ndim == 1 or isinstance(X, list)
                and not hasattr(X[0], "__len__"))

    if has_one_axis(X):
        X = [X]
    if Y is None:
        X, Y = [[]] * len(X), X
    elif has_one_axis(Y):
        Y = [Y]
    if len(X) != len(Y):
        X = X * len(Y)
    axes.cla()
    for x, y, fmt in zip(X, Y, fmts):
        if len(x):
            axes.plot(x, y, fmt)
        else:
            axes.plot(y, fmt)
    set_axes(axes, xlabel, ylabel, xlim, ylim, xscale, yscale, legend)

x = np.arange(0, 3, 0.1)
plot(x, [f(x), 2 * x - 3], 'x', 'f(x)', legend=['f(x)', 'Tangent line (x=1)'])



正态分布绘图
# 再次使用numpy进行可视化
x = np.arange(-7, 7, 0.01)

# 均值和标准差对
params = [(0, 1), (0, 2), (3, 1)]
d2l.plot(x, [normal(x, mu, sigma) for mu, sigma in params], xlabel='x',
         ylabel='p(x)', figsize=(4.5, 2.5),
         legend=[f'mean {mu}, std {sigma}' for mu, sigma in params])

概率
fair_probs = torch.ones([6]) / 6   这是概率分布
counts =multinomial.Multinomial(10, fair_probs).sample()这是多项式中抽样次数，概率分布，生成样本，形成的是一个len6张量频率
counts/10频率到概率


概率论
样本空间sample or outcome space中每个元素都是结果。
事件是结果的各种集合
联合概率，条件概率，贝叶斯公式，边际化...马尔可夫链，友元图

案例

线性回归模型训练

def sgd(params, lr, batch_size):  #@save
    """小批量随机梯度下降"""
    with torch.no_grad():
        for param in params:
            param -= lr * param.grad / batch_size
            param.grad.zero_()
torch.no_grad() 创建一个上下文管理器，该管理器中的计算操作不会被记录用于自动求导。
对于参数列表中的每个参数 param，执行以下操作：
param -= lr * param.grad / batch_size：使用梯度下降更新参数。根据梯度 param.grad、学习率 lr 和批量大小 batch_size，更新参数 param 的值。
param.grad.zero_()：将参数的梯度清零，以便在下一次计算梯度时不受之前梯度的影响。

依靠API