Skip to content

function

Source code in tinybig/module/base_function.py 
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
class function:
    def __init__(self, name: str = 'base_function', device: str = 'cpu', *args, **kwargs):
        self.name = name
        self.device = device

    def __call__(self, *args, **kwargs):
        self.forward(*args, **kwargs)

    def to_config(self):
        class_name = f"{self.__class__.__module__}.{self.__class__.__name__}"
        attributes = {attr: getattr(self, attr) for attr in self.__dict__}
        return {
            "function_class": class_name,
            "function_parameters": attributes
        }

    @staticmethod
    def func_x(x, functions, device: str = 'cpu'):
        """
        The function execution to the input data.

        It applies the list of functions to the input vector and returns the calculation results.

        This method will be extensively called for handeling the data processing functions in the
        expansion functions, RPN head and remainder functions in tinyBIG.

        * preprocess_functions in expansion functions
        * postprocess_functions in expansion functions
        * output_process_functions in rpn heads
        * activation_functions in remainder functions

        Parameters
        ----------
        x: torch.Tensor
            The input data vector.
        functions: list | tuple | callable
            The functions to be applied to the input vector. The function can be callable functions,
            string names of the functions, the complete class descriptions of the functions, etc.
        device: str, default = 'cpu'
            The device to perform the function on the input vector.

        Returns
        -------
        torch.Tensor
            The processed input vector by these functions.
        """
        if functions is None or ((isinstance(functions, list) or isinstance(functions, tuple)) and len(functions) == 0):
            return x
        elif isinstance(functions, list) or isinstance(functions, tuple):
            for f in functions:
                if callable(f):
                    x = f(x)
                elif type(f) is str:
                    x = function.str_func_x(x=x, func=f, device=device)
            return x
        else:
            if callable(functions):
                return functions(x)
            elif type(functions) is str:
                return function.str_func_x(x=x, func=functions, device=device)

    @staticmethod
    def str_func_x(x, func: str | Callable, device='cpu', *args, **kwargs):
        """
        Function recognition from their string names or string class descriptions.

        It recognizes the data processing functions from their names or class description in strings,
        e.g., "layer_norm" or "torch.nn.functional.layer_norm".

        Since these functions can be very diverse, whose definitions are also very different,
        it makes it very challenging to process them based on their string descriptions.
        This method can process some basic functions, e.g., activation functions, and normalization functions.
        For the functions that are not implemented in this method, the users may consider to extend the method
        to handle more complex input functions.

        Parameters
        ----------
        x: torch.Tensor
            The input data vector.
        func: str
            The string description of the functoin name or class.
        device: str, default = 'cpu'
            The device to host and apply the recognized functions.

        Returns
        -------
        torch.Tensor
            The processed input data vector by the recognized functions.
        """
        if func is None:
            return x
        elif callable(func):
            # --------------------------
            if func in [F.sigmoid, F.relu, F.leaky_relu, F.tanh, F.softplus, F.silu, F.celu, F.gelu]:
                return func(x)
            # --------------------------
            # dropout functions
            elif func in [
                F.dropout
            ]:
                # --------------------------
                if 'p' in kwargs:
                    p = kwargs['p']
                else:
                    p = 0.5
                # --------------------------
                if func in [F.dropout]:
                    return func(x, p=p)
                else:
                    return func(p=p)(x)
            # --------------------------
            # layer_norm functions
            elif func in [F.layer_norm]:
                # --------------------------
                if 'normalized_shape' in kwargs:
                    normalized_shape = kwargs['normalized_shape']
                else:
                    normalized_shape = [x.size(-1)]
                # --------------------------
                if func in [F.layer_norm]:
                    return func(x, normalized_shape=normalized_shape)
                else:
                    return func(normalized_shape=normalized_shape)(x)
                # --------------------------
            # --------------------------
            # batch_norm functions
            elif func in [
                torch.nn.BatchNorm1d, torch.nn.BatchNorm2d, torch.nn.BatchNorm3d,
                torch.nn.modules.batchnorm.BatchNorm1d,
                torch.nn.modules.batchnorm.BatchNorm2d,
                torch.nn.modules.batchnorm.BatchNorm3d
            ]:
                # --------------------------
                if 'num_features' in kwargs:
                    num_features = kwargs['num_features']
                else:
                    num_features = x.size(-1)
                # ---------------------------
                return func(num_features=num_features, device=device)(x)
            # --------------------------
            # other functions
            elif func in [
                torch.exp,
            ]:
                return func(x)
            # --------------------------
            else:
                warnings.warn(
                    'input function {} not recognized, the original input x will be returned by default...'.format(
                        func),
                    UserWarning)
                return x
        # ------------------------------
        # All functions from configs will convert from str to object first
        elif type(func) is str:
            try:
                if '.' in func:
                    func = config.get_obj_from_str(func)
                else:
                    func = config.get_obj_from_str("torch.nn.functional.{}".format(func.lower()))
            except:
                raise ValueError(
                    'function {} does\'t belong to "torch.nn.functional.", please provide the complete callable function path, such as "torch.nn.functional.sigmoid..."'.format(
                        func))
            return function.str_func_x(x, func, device=device, *args, **kwargs)
        else:
            warnings.warn('input function not recognized, the original input x will be returned by default...',
                          UserWarning)
            return x

    @staticmethod
    def string_to_function(formula, variable):
        """
        Formula recognition from strings.

        It recognizes and returns the formula and variables from strings via the sympy package.

        Parameters
        ----------
        formula: str
            The function formula as a string.
        variable: list
            The list of the variables involved in the formula.

        Returns
        -------
        sympy.FunctionClass
            The recognized function of the input formula.
        """
        # Define the symbol
        var = sp.symbols(variable)

        # Parse the formula string into a sympy expression
        expression = sp.sympify(formula)

        # Convert the sympy expression to a lambda function
        func = sp.lambdify(var, expression, 'numpy')

        return func

    @staticmethod
    def functions_to_configs(functions: list | tuple | Callable, class_name: str = 'function_class', parameter_name: str = 'function_parameters'):
        if functions is None:
            return None
        elif isinstance(functions, Callable):
            func_class_name = f"{functions.__class__.__module__}.{functions.__class__.__name__}"
            func_parameters = {attr: getattr(functions, attr) for attr in functions.__dict__}
            return {
                class_name: func_class_name,
                parameter_name: func_parameters
            }
        else:
            return [
                function.functions_to_configs(func) for func in functions
            ]

    @abstractmethod
    def forward(self, *args, **kwargs):
        pass

func_x(x, functions, device='cpu') staticmethod

The function execution to the input data.

It applies the list of functions to the input vector and returns the calculation results.

This method will be extensively called for handeling the data processing functions in the expansion functions, RPN head and remainder functions in tinyBIG.

  • preprocess_functions in expansion functions
  • postprocess_functions in expansion functions
  • output_process_functions in rpn heads
  • activation_functions in remainder functions

Parameters:

Name Type Description Default
x

The input data vector.

 required
functions

The functions to be applied to the input vector. The function can be callable functions, string names of the functions, the complete class descriptions of the functions, etc.

 required
device str

The device to perform the function on the input vector.

 'cpu'

Returns:

Type Description
Tensor

The processed input vector by these functions.
Source code in tinybig/module/base_function.py 
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
@staticmethod
def func_x(x, functions, device: str = 'cpu'):
    """
    The function execution to the input data.

    It applies the list of functions to the input vector and returns the calculation results.

    This method will be extensively called for handeling the data processing functions in the
    expansion functions, RPN head and remainder functions in tinyBIG.

    * preprocess_functions in expansion functions
    * postprocess_functions in expansion functions
    * output_process_functions in rpn heads
    * activation_functions in remainder functions

    Parameters
    ----------
    x: torch.Tensor
        The input data vector.
    functions: list | tuple | callable
        The functions to be applied to the input vector. The function can be callable functions,
        string names of the functions, the complete class descriptions of the functions, etc.
    device: str, default = 'cpu'
        The device to perform the function on the input vector.

    Returns
    -------
    torch.Tensor
        The processed input vector by these functions.
    """
    if functions is None or ((isinstance(functions, list) or isinstance(functions, tuple)) and len(functions) == 0):
        return x
    elif isinstance(functions, list) or isinstance(functions, tuple):
        for f in functions:
            if callable(f):
                x = f(x)
            elif type(f) is str:
                x = function.str_func_x(x=x, func=f, device=device)
        return x
    else:
        if callable(functions):
            return functions(x)
        elif type(functions) is str:
            return function.str_func_x(x=x, func=functions, device=device)

str_func_x(x, func, device='cpu', *args, **kwargs) staticmethod

Function recognition from their string names or string class descriptions.

It recognizes the data processing functions from their names or class description in strings, e.g., "layer_norm" or "torch.nn.functional.layer_norm".

Since these functions can be very diverse, whose definitions are also very different, it makes it very challenging to process them based on their string descriptions. This method can process some basic functions, e.g., activation functions, and normalization functions. For the functions that are not implemented in this method, the users may consider to extend the method to handle more complex input functions.

Parameters:

Name Type Description Default
x

The input data vector.

 required
func str | Callable

The string description of the functoin name or class.

 required
device

The device to host and apply the recognized functions.

 'cpu'

Returns:

Type Description
Tensor

The processed input data vector by the recognized functions.
Source code in tinybig/module/base_function.py 
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
@staticmethod
def str_func_x(x, func: str | Callable, device='cpu', *args, **kwargs):
    """
    Function recognition from their string names or string class descriptions.

    It recognizes the data processing functions from their names or class description in strings,
    e.g., "layer_norm" or "torch.nn.functional.layer_norm".

    Since these functions can be very diverse, whose definitions are also very different,
    it makes it very challenging to process them based on their string descriptions.
    This method can process some basic functions, e.g., activation functions, and normalization functions.
    For the functions that are not implemented in this method, the users may consider to extend the method
    to handle more complex input functions.

    Parameters
    ----------
    x: torch.Tensor
        The input data vector.
    func: str
        The string description of the functoin name or class.
    device: str, default = 'cpu'
        The device to host and apply the recognized functions.

    Returns
    -------
    torch.Tensor
        The processed input data vector by the recognized functions.
    """
    if func is None:
        return x
    elif callable(func):
        # --------------------------
        if func in [F.sigmoid, F.relu, F.leaky_relu, F.tanh, F.softplus, F.silu, F.celu, F.gelu]:
            return func(x)
        # --------------------------
        # dropout functions
        elif func in [
            F.dropout
        ]:
            # --------------------------
            if 'p' in kwargs:
                p = kwargs['p']
            else:
                p = 0.5
            # --------------------------
            if func in [F.dropout]:
                return func(x, p=p)
            else:
                return func(p=p)(x)
        # --------------------------
        # layer_norm functions
        elif func in [F.layer_norm]:
            # --------------------------
            if 'normalized_shape' in kwargs:
                normalized_shape = kwargs['normalized_shape']
            else:
                normalized_shape = [x.size(-1)]
            # --------------------------
            if func in [F.layer_norm]:
                return func(x, normalized_shape=normalized_shape)
            else:
                return func(normalized_shape=normalized_shape)(x)
            # --------------------------
        # --------------------------
        # batch_norm functions
        elif func in [
            torch.nn.BatchNorm1d, torch.nn.BatchNorm2d, torch.nn.BatchNorm3d,
            torch.nn.modules.batchnorm.BatchNorm1d,
            torch.nn.modules.batchnorm.BatchNorm2d,
            torch.nn.modules.batchnorm.BatchNorm3d
        ]:
            # --------------------------
            if 'num_features' in kwargs:
                num_features = kwargs['num_features']
            else:
                num_features = x.size(-1)
            # ---------------------------
            return func(num_features=num_features, device=device)(x)
        # --------------------------
        # other functions
        elif func in [
            torch.exp,
        ]:
            return func(x)
        # --------------------------
        else:
            warnings.warn(
                'input function {} not recognized, the original input x will be returned by default...'.format(
                    func),
                UserWarning)
            return x
    # ------------------------------
    # All functions from configs will convert from str to object first
    elif type(func) is str:
        try:
            if '.' in func:
                func = config.get_obj_from_str(func)
            else:
                func = config.get_obj_from_str("torch.nn.functional.{}".format(func.lower()))
        except:
            raise ValueError(
                'function {} does\'t belong to "torch.nn.functional.", please provide the complete callable function path, such as "torch.nn.functional.sigmoid..."'.format(
                    func))
        return function.str_func_x(x, func, device=device, *args, **kwargs)
    else:
        warnings.warn('input function not recognized, the original input x will be returned by default...',
                      UserWarning)
        return x

string_to_function(formula, variable) staticmethod

Formula recognition from strings.

It recognizes and returns the formula and variables from strings via the sympy package.

Parameters:

Name Type Description Default
formula

The function formula as a string.

 required
variable

The list of the variables involved in the formula.

 required

Returns:

Type Description
FunctionClass

The recognized function of the input formula.
Source code in tinybig/module/base_function.py 
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
@staticmethod
def string_to_function(formula, variable):
    """
    Formula recognition from strings.

    It recognizes and returns the formula and variables from strings via the sympy package.

    Parameters
    ----------
    formula: str
        The function formula as a string.
    variable: list
        The list of the variables involved in the formula.

    Returns
    -------
    sympy.FunctionClass
        The recognized function of the input formula.
    """
    # Define the symbol
    var = sp.symbols(variable)

    # Parse the formula string into a sympy expression
    expression = sp.sympify(formula)

    # Convert the sympy expression to a lambda function
    func = sp.lambdify(var, expression, 'numpy')

    return func