PyTorch

PyTorch è un modulo esterno del linguaggio Python con diverse funzioni dedicate al machine learning e al deep learning.

La libreria PyTorch ha le stesse funzionalità di Numpy per quanto riguarda l'elaborazione degli array multidimensionali ma è molto più ampia e potente.

E' particolarmente utile per elaborare i tensori usando l'accelerazione delle GPU delle schede grafiche. Spesso uso PyTorch nell'elaborazione delle reti neurali ed è molto veloce.

Come installare PyTorch

Per usare PyTorch devo prima installarlo in Python tramite il comando pip.

pip install torch

Se non funziona, posso sempre scaricare e installare il modulo manualmente

Ad esempio, per installare PyTorch su Python 3.7

pip3 install https://download.pytorch.org/whl/cpu/torch-1.0.1-cp37-cp37m-win_amd64.whl
pip3 install torchvision

Ovviamente, il modulo da scaricare cambia a seconda della versione di Python.

Per qualsiasi info aggiornata si può consultare la pagina ufficiale di PyTorch.

Dopo averlo installato, posso importare il modulo PyTorch nell'interprete.

>>> import torch

E così via

Le funzioni e i metodi di PyTorch

Ecco l'elenco completo delle funzioni e dei metodi di PyTorch

• Argument
• ArgumentSpec
• Block
• BoolType
• ByteStorage
• ByteTensor
• CharStorage
• CharTensor
• Code
• CompleteArgumentSpec
• DoubleStorage
• DoubleTensor
• DynamicType
• ExecutionPlanState
• FatalError
• FloatStorage
• FloatTensor
• FloatType
• FunctionSchema
• Future
• Generator
• Gradient
• Graph
• GraphExecutor
• GraphExecutorState
• HalfStorage
• HalfStorageBase
• HalfTensor
• IODescriptor
• IS_CONDA
• IntStorage
• IntTensor
• IntType
• JITException
• ListType
• LongStorage
• LongTensor
• Node
• NumberType
• PyTorchFileReader
• PyTorchFileWriter
• ScriptMethod
• ScriptModule
• ShortStorage
• ShortTensor
• Size
• Storage
• Tensor
• TracingState
• TupleType
• Type
• Use
• Value
• _C
• _StorageBase
• __all__
• __builtins__
• __cached__
• __doc__
• __file__
• __loader__
• __name__
• __package__
• __path__
• __spec__
• __version__
• _argmax
• _argmin
• _baddbmm_mkl_
• _cast_Byte
• _cast_Char
• _cast_Double
• _cast_Float
• _cast_Half
• _cast_Int
• _cast_Long
• _cast_Short
• _convolution
• _convolution_nogroup
• _copy_same_type_
• _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
• _dim_arange
• _dirichlet_grad
• _embedding_bag
• _fft_with_size
• _fused_dropout
• _import_dotted_name
• _jit_internal
• _log_softmax
• _log_softmax_backward_data
• _masked_scale
• _np
• _ops
• _pack_padded_sequence
• _pad_packed_sequence
• _promote_types
• _reshape_from_tensor
• _s_copy_from
• _s_where
• _shape_as_tensor
• _six
• _softmax
• _softmax_backward_data
• _sparse_addmm
• _sparse_mm
• _sparse_sum
• _standard_gamma
• _standard_gamma_grad
• _storage_classes
• _string_classes
• _tensor_classes
• _tensor_str
• _thnn
• _trilinear
• _unique
• _unique_dim
• _utils
• _utils_internal
• _weight_norm
• _weight_norm_cuda_interface
• abs
• abs_
• acos
• acos_
• adaptive_avg_pool1d
• adaptive_max_pool1d
• add
• add_extra_dll_dir
• addbmm
• addcdiv
• addcmul
• addmm
• addmv
• addmv_
• addr
• affine_grid_generator
• all
• allclose
• alpha_dropout
• alpha_dropout_
• any
• arange
• argmax
• argmin
• argsort
• as_strided
• as_strided_
• as_tensor
• asin
• asin_
• atan
• atan2
• atan_
• autograd
• avg_pool1d
• backends
• baddbmm
• bartlett_window
• batch_norm
• bernoulli
• bilinear
• binary_cross_entropy_with_logits
• bincount
• blackman_window
• bmm
• broadcast_tensors
• btrifact
• btrifact_with_info
• btrisolve
• btriunpack
• cat
• ceil
• ceil_
• celu
• celu_
• chain_matmul
• cholesky
• chunk
• clamp
• clamp_
• clamp_max
• clamp_max_
• clamp_min
• clamp_min_
• clone
• compiled_with_cxx11_abi
• complex128
• complex32
• complex64
• constant_pad_nd
• conv1d
• conv2d
• conv3d
• conv_tbc
• conv_transpose1d
• conv_transpose2d
• conv_transpose3d
• convolution
• cos
• cos_
• cosh
• cosh_
• cosine_embedding_loss
• cosine_similarity
• cross
• ctc_loss
• cuda
• cudnn_affine_grid_generator
• cudnn_batch_norm
• cudnn_convolution
• cudnn_convolution_transpose
• cudnn_grid_sampler
• cudnn_is_acceptable
• cumprod
• cumsum
• default_generator
• det
• detach
• detach_
• device
• diag
• diag_embed
• diagflat
• diagonal
• digamma
• dist
• distributed
• distributions
• div
• dll_paths
• dot
• double
• dropout
• dropout_
• dsmm
• dtype
• eig
• einsum
• embedding
• embedding_bag
• embedding_renorm_
• empty
• empty_like
• empty_strided
• enable_grad
• eq
• equal
• erf
• erf_
• erfc
• erfc_
• erfinv
• exp
• exp_
• expm1
• expm1_
• eye
• feature_alpha_dropout
• feature_alpha_dropout_
• feature_dropout
• feature_dropout_
• fft
• fill_
• finfo
• flatten
• flip
• float
• float16
• float32
• float64
• floor
• floor_
• fmod
• fork
• frac
• frobenius_norm
• from_numpy
• full
• full_like
• functional
• gather
• ge
• gels
• geqrf
• ger
• gesv
• get_default_dtype
• get_device
• get_file_path
• get_num_threads
• get_nvToolsExt_path
• get_rng_state
• grid_sampler
• grid_sampler_2d
• grid_sampler_3d
• group_norm
• gru
• gru_cell
• gt
• half
• hamming_window
• hann_window
• hardshrink
• has_cudnn
• has_lapack
• has_mkl
• hinge_embedding_loss
• histc
• hsmm
• hspmm
• ifft
• iinfo
• import_ir_module
• import_ir_module_from_buffer
• index_put
• index_put_
• index_select
• initial_seed
• instance_norm
• int
• int16
• int32
• int64
• int8
• inverse
• irfft
• is_anomaly_enabled
• is_complex
• is_distributed
• is_floating_point
• is_grad_enabled
• is_nonzero
• is_same_size
• is_signed
• is_storage
• is_tensor
• isclose
• isfinite
• isinf
• isnan
• jit
• kl_div
• kthvalue
• layer_norm
• layout
• le
• lerp
• lgamma
• linspace
• load
• log
• log10
• log10_
• log1p
• log1p_
• log2
• log2_
• log_
• log_softmax
• logdet
• logspace
• logsumexp
• long
• lstm
• lstm_cell
• lt
• manual_seed
• margin_ranking_loss
• masked_select
• matmul
• matrix_power
• matrix_rank
• max
• max_pool1d_with_indices
• mean
• median
• merge_type_from_type_comment
• meshgrid
• min
• miopen_batch_norm
• miopen_convolution
• miopen_convolution_transpose
• mkldnn_convolution
• mkldnn_convolution_backward_weights
• mm
• mode
• mul
• multinomial
• multiprocessing
• mv
• mvlgamma
• name
• narrow
• native_batch_norm
• native_clone
• native_norm
• native_pow
• native_resize_as_
• native_zero_
• ne
• neg
• nn
• no_grad
• nonzero
• norm
• norm_except_dim
• normal
• nuclear_norm
• numel
• ones
• ones_like
• onnx
• ops
• optim
• orgqr
• ormqr
• os
• p
• pairwise_distance
• parse_type_comment
• pdist
• pin_memory
• pinverse
• pixel_shuffle
• platform
• poisson
• polygamma
• potrf
• potri
• potrs
• pow
• prelu
• prepare_multiprocessing_environment
• prod
• pstrf
• py_dll_path
• qr
• rand
• rand_like
• randint
• randint_like
• randn
• randn_like
• random
• randperm
• range
• reciprocal
• register_batch_operator
• relu
• relu_
• remainder
• renorm
• reshape
• resize_as_
• rfft
• rnn_relu
• rnn_relu_cell
• rnn_tanh
• rnn_tanh_cell
• roll
• rot90
• round
• round_
• rrelu
• rrelu_
• rsqrt
• rsqrt_
• rsub
• s_copy_
• s_native_addmm
• s_native_addmm_
• saddmm
• save
• select
• selu
• selu_
• serialization
• set_anomaly_enabled
• set_default_dtype
• set_default_tensor_type
• set_flush_denormal
• set_grad_enabled
• set_num_threads
• set_printoptions
• set_rng_state
• short
• sigmoid
• sigmoid_
• sign
• sin
• sin_
• sinh
• sinh_
• slogdet
• smm
• softmax
• sort
• sparse
• sparse_coo
• sparse_coo_tensor
• split
• split_with_sizes
• spmm
• sqrt
• sqrt_
• squeeze
• sspaddmm
• stack
• std
• stft
• storage
• strided
• sub
• sum
• svd
• symeig
• sys
• t
• take
• tan
• tan_
• tanh
• tanh_
• tensor
  crea un tensore, una matrice o un vettore
• tensordot
• testing
• th_dll_path
• threshold
• threshold_
• to_batch_graph
• topk
• torch
• trace
• transpose
• tril
• triplet_margin_loss
• triu
• trtrs
• trunc
• trunc_
• typename
• uint8
• unbind
• unique
• unsqueeze
• utils
• var
• version
• wait
• where
• zero_
• zeros
• zeros_like