pyerualjetwork 5.37__py3-none-any.whl → 5.40a0__py3-none-any.whl

pyerualjetwork/model_ops.py

@@ -0,0 +1,692 @@
+""" 
+
+
+Model Operations
+=======================
+This module hosts functions for handling all operational processes related to models, including:
+
+- Saving and loading models
+- Making predictions from memory
+- Making predictions from storage
+- Retrieving model weights
+- Retrieving model activation functions
+- Retrieving model accuracy
+- Running the model in reverse (applicable to PLAN models)
+
+Module functions:
+-----------------
+- get_model_template()
+- save_model()
+- load_model()
+- predict_from_storage()
+- predict_from_memory()
+- reverse_predict_from_storage()
+- reverse_predict_from_memory()
+
+Examples: https://github.com/HCB06/PyerualJetwork/tree/main/Welcome_to_PyerualJetwork/ExampleCodes
+
+PyerualJetwork document: https://github.com/HCB06/PyerualJetwork/blob/main/Welcome_to_PyerualJetwork/PYERUALJETWORK_USER_MANUEL_AND_LEGAL_INFORMATION(EN).pdf
+
+- Creator: Hasan Can Beydili
+- YouTube: https://www.youtube.com/@HasanCanBeydili
+- Linkedin: https://www.linkedin.com/in/hasan-can-beydili-77a1b9270/
+- Instagram: https://www.instagram.com/canbeydilj
+- Contact: tchasancan@gmail.com
+"""
+
+import numpy as np
+import cupy as cp
+from colorama import Fore, Style
+from datetime import datetime
+import pickle
+from scipy import io
+import scipy.io as sio
+import pandas as pd
+from collections import namedtuple
+
+def get_model_template():
+    
+    Model = namedtuple("Model", [
+    "weights",
+    "predictions",
+    "accuracy",
+    "activations",
+    "scaler_params",
+    "softmax_predictions",
+    "model_type",
+    "weights_type",
+    "weights_format",
+    "model_version",
+    "model_df",
+    "activation_potentiation"
+    ])
+    
+    return Model
+
+
+def save_model(model,
+               model_name,
+               model_path='',
+               weights_type='npy',
+               weights_format='raw',
+               show_architecture=False,
+               show_info=True
+               ):
+
+    """
+    Function to save a potentiation learning artificial neural network model.
+    Args:
+
+        model (tuple): Trained model.
+
+        model_name: (str): Name of the model.
+        
+        model_path: (str): Path where the model will be saved. For example: C:/Users/beydili/Desktop/denemePLAN/ default: ''
+
+        weights_type: (str): Type of weights to save (options: 'txt', 'pkl', 'npy', 'mat'). default: 'npy'
+        
+        weights_format: (str): Format of the weights (options: 'f', 'raw'). default: 'raw'
+        
+        show_architecture: (bool): It draws model architecture. True or False. Default: False. NOTE! draw architecture only works for PLAN models. Not MLP models for now, but it will be.
+        
+        show_info: (bool): Prints model details into console. default: True
+
+    Returns:
+        No return.
+    """
+    
+    from .cpu.visualizations import draw_model_architecture
+    from . import __version__
+
+    model_type = model.model_type
+    activations = model.activations
+    activation_potentiation = model.activation_potentiation
+    scaler_params = model.scaler_params
+    W = model.weights
+    acc = model.accuracy
+
+    if model_type != 'PLAN' and model_type != 'MLP' and model_type != 'PTNN':
+        raise ValueError("model_type parameter must be 'PLAN', 'MLP' or 'PTNN'.")
+
+    if model_type == 'PTNN' and activation_potentiation == []:
+        raise ValueError('PTNN models need extra activation_potentiation parameter.')
+
+    if isinstance(activations, str):
+        activations = [activations]
+    else:
+        activations = [item if isinstance(item, list) else [item] for item in activations]
+
+    activations = activations.copy()
+
+    if model_type == 'PTNN':
+        if isinstance(activation_potentiation, str):
+            activation_potentiation = [activation_potentiation]
+        else:
+            activation_potentiation = [item if isinstance(item, list) else [item] for item in activation_potentiation]
+
+        activation_potentiation = activation_potentiation.copy()
+
+    if acc != None:
+        acc= float(acc)
+
+    if weights_type != 'txt' and weights_type != 'npy' and weights_type != 'mat' and weights_type != 'pkl':
+        raise ValueError(Fore.RED + "ERROR: Save Weight type (File Extension) Type must be 'txt' or 'npy' or 'mat' or 'pkl' from: save_model" + Style.RESET_ALL)
+
+    if weights_format != 'd' and weights_format != 'f' and weights_format != 'raw':
+        raise ValueError(Fore.RED + "ERROR: Weight Format Type must be 'd' or 'f' or 'raw' from: save_model" + Style.RESET_ALL)
+        
+    NeuronCount = []
+    SynapseCount = []
+
+    if model_type == 'PLAN':
+        class_count = W.shape[0]
+
+        try:
+            NeuronCount.append(np.shape(W)[1])
+            NeuronCount.append(np.shape(W)[0])
+            SynapseCount.append(np.shape(W)[0] * np.shape(W)[1])
+        except AttributeError as e:
+            raise AttributeError(Fore.RED + "ERROR: W does not have a shape attribute. Check if W is a valid matrix." + Style.RESET_ALL) from e
+        except IndexError as e:
+            raise IndexError(Fore.RED + "ERROR: W has an unexpected shape format. Ensure it has two dimensions." + Style.RESET_ALL) from e
+        except (TypeError, ValueError) as e:
+            raise TypeError(Fore.RED + "ERROR: W is not a valid numeric matrix." + Style.RESET_ALL) from e
+        except Exception as e:
+            raise RuntimeError(Fore.RED + f"ERROR: An unexpected error occurred in save_model: {e}" + Style.RESET_ALL) from e
+
+    elif model_type == 'MLP' or model_type == 'PTNN':
+
+        class_count = W[-1].shape[0]
+        
+        NeuronCount.append(np.shape(W[0])[1])
+
+        for i in range(len(W)):
+            try:
+                    NeuronCount.append(np.shape(W[i])[0])
+                    SynapseCount.append(np.shape(W[i])[0] * np.shape(W[i])[1])
+            except AttributeError as e:
+                raise AttributeError(Fore.RED + "ERROR: W does not have a shape attribute. Check if W is a valid matrix." + Style.RESET_ALL) from e
+            except IndexError as e:
+                raise IndexError(Fore.RED + "ERROR: W has an unexpected shape format. Ensure it has two dimensions." + Style.RESET_ALL) from e
+            except (TypeError, ValueError) as e:
+                raise TypeError(Fore.RED + "ERROR: W is not a valid numeric matrix." + Style.RESET_ALL) from e
+            except Exception as e:
+                raise RuntimeError(Fore.RED + f"ERROR: An unexpected error occurred in save_model: {e}" + Style.RESET_ALL) from e
+
+    
+        SynapseCount.append(' ')
+        
+        activations.append('')
+        activations.insert(0, '')
+
+    if len(activations) == 1 and model_type == 'PLAN':
+        activations = [activations]
+        activations.append('')
+
+    if model_type == 'PTNN':
+        if len(activations) > len(activation_potentiation):
+            for i in range(len(activations) - len(activation_potentiation)):
+                activation_potentiation.append('')
+
+        if len(activation_potentiation) > len(activations):
+            for i in range(len(activation_potentiation) - len(activations)):
+                activations.append('')
+
+    if len(activations) > len(NeuronCount):
+        for i in range(len(activations) - len(NeuronCount)):
+            NeuronCount.append('')
+        
+    if len(activations) > len(SynapseCount):
+        for i in range(len(activations) - len(SynapseCount)):
+            SynapseCount.append('')
+
+    if scaler_params != None:
+
+        if len(scaler_params) > len(activations):
+
+            activations += ['']
+
+        elif len(activations) > len(scaler_params):
+
+            for i in range(len(activations) - len(scaler_params)):
+
+                scaler_params.append(' ')
+
+    data = {'MODEL NAME': model_name,
+            'MODEL TYPE': model_type,
+            'CLASS COUNT': class_count,
+            'NEURON COUNT': NeuronCount,
+            'SYNAPSE COUNT': SynapseCount,
+            'VERSION': __version__,
+            'ACCURACY': acc,
+            'SAVE DATE': datetime.now(),
+            'WEIGHTS TYPE': weights_type,
+            'WEIGHTS FORMAT': weights_format,
+            'STANDARD SCALER': scaler_params,
+            'ACTIVATION FUNCTIONS': activations,
+            'ACTIVATION POTENTIATION': activation_potentiation
+            }
+
+    df = pd.DataFrame(data)
+    df.to_pickle(model_path + model_name + '.pkl')
+
+    try:
+
+        if weights_type == 'txt' and weights_format == 'f':
+
+                np.savetxt(model_path + model_name + f'_weights.txt',  W, fmt='%f')
+
+        if weights_type == 'txt' and weights_format == 'raw':
+
+                np.savetxt(model_path + model_name + f'_weights.txt',  W)
+
+        ###
+
+        
+        if weights_type == 'pkl' and weights_format == 'f':
+
+            with open(model_path + model_name + f'_weights.pkl', 'wb') as f:
+                pickle.dump(W.astype(float), f)
+
+        if weights_type == 'pkl' and weights_format =='raw':
+        
+            with open(model_path + model_name + f'_weights.pkl', 'wb') as f:
+                pickle.dump(W, f)
+
+        ###
+
+        if weights_type == 'npy' and weights_format == 'f':
+
+                np.save(model_path + model_name + f'_weights.npy',  W, W.astype(float))
+
+        if weights_type == 'npy' and weights_format == 'raw':
+
+                np.save(model_path + model_name + f'_weights.npy',  W)
+
+        ###
+
+        if weights_type == 'mat' and weights_format == 'f':
+
+                w = {'w': W.astype(float)}
+                io.savemat(model_path + model_name + f'_weights.mat', w)
+
+        if weights_type == 'mat' and weights_format == 'raw':
+                
+                w = {'w': W}
+                io.savemat(model_path + model_name + f'_weights.mat', w)
+
+
+    except OSError as e:
+        raise OSError(Fore.RED + f"ERROR: An OSError error occurred in save_model at saving weights. Maybe model name or path or administration issue: {e}" + Style.RESET_ALL) from e
+
+    if show_info:
+        print(df)
+    
+        message = (
+            Fore.GREEN + "Model Saved Successfully\n" +
+            Fore.MAGENTA + "Don't forget, if you want to load model: model log file and weight files must be in the same directory." +
+            Style.RESET_ALL
+        )
+        
+        print(message)
+
+    if show_architecture:
+        draw_model_architecture(model_name=model_name, model_path=model_path)
+
+
+
+def load_model(model_name,
+               model_path,
+               ):
+    """
+   Function to load a potentiation learning model.
+
+   Args:
+    model_name (str): Name of the model.
+    
+    model_path (str): Path where the model is saved.
+
+   Returns:
+    tuple(model): Weights, None, accuracy, activations, scaler_params, None, model_type, weight_type, weight_format, device_version, (list[df_elements])=Pandas DataFrame of the model, activation_potentiation
+    """
+
+    from . import __version__
+
+    try:
+
+         df = pd.read_pickle(model_path + model_name + '.pkl')
+
+    except OSError as e:
+        raise OSError(Fore.RED + f"ERROR: An OSError error occurred in load_model at loading model params. Maybe model name or path or administration issue: {e}" + Style.RESET_ALL) from e
+
+
+
+    scaler_params = df['STANDARD SCALER'].tolist()
+    
+    try:
+        if scaler_params[0] == None:
+            scaler_params = scaler_params[0]
+
+    except:
+        scaler_params = [item for item in scaler_params if isinstance(item, np.ndarray)]
+
+     
+    model_name = str(df['MODEL NAME'].iloc[0])
+    model_type = str(df['MODEL TYPE'].iloc[0])
+    WeightType = str(df['WEIGHTS TYPE'].iloc[0])
+    WeightFormat = str(df['WEIGHTS FORMAT'].iloc[0])
+    acc = str(df['ACCURACY'].iloc[0])
+
+    activations = list(df['ACTIVATION FUNCTIONS'])
+    activations = [x for x in activations if not (isinstance(x, float) and np.isnan(x))]
+    activations = [item for item in activations if item != ''] 
+
+    activation_potentiation = list(df['ACTIVATION POTENTIATION'])
+    activation_potentiation = [x for x in activation_potentiation if not (isinstance(x, float) and np.isnan(x))]
+    activation_potentiation = [item for item in activation_potentiation if item != ''] 
+
+    device_version = __version__
+
+    try:
+        model_version = str(df['VERSION'].iloc[0])
+        if model_version != device_version:
+            message = (
+            Fore.MAGENTA + f"WARNING: Your PyerualJetwork version({device_version}) is different from this model's version({model_version}).\nIf you have a performance issue, please install this model version. Use this: pip install pyerualjetwork=={model_version} or look issue_solver module." +
+            Style.RESET_ALL
+        )
+            print(message)
+        
+    except:
+        pass # Version check only in >= 5.0.2
+
+    if model_type == 'MLP' or model_type == 'PTNN': allow_pickle = True
+    else: allow_pickle = False
+
+    try:
+        if WeightType == 'txt':
+                W = (np.loadtxt(model_path + model_name + f'_weights.txt'))
+        elif WeightType == 'npy':
+                W = (np.load(model_path + model_name + f'_weights.npy', allow_pickle=allow_pickle))
+        elif WeightType == 'mat':
+                W = (sio.loadmat(model_path + model_name + f'_weights.mat'))
+        elif WeightType == 'pkl':
+            with open(model_path + model_name + f'_weights.pkl', 'rb') as f:
+                W = pickle.load(f)
+        else:
+
+            raise ValueError(
+                Fore.RED + "Incorrect weight type value. Value must be 'txt', 'npy', 'pkl' or 'mat' from: load_model." + Style.RESET_ALL)
+
+    except OSError as e:
+        raise OSError(Fore.RED + f"ERROR: An OSError error occurred in load_model at loading weights. Maybe model name or path or administration issue: {e}" + Style.RESET_ALL) from e
+
+
+    if WeightType == 'mat':
+        W = W['w']
+
+    Model = get_model_template()
+
+    model = Model(
+    W,
+    None,
+    acc,
+    activations,
+    scaler_params,
+    None,
+    model_type,
+    WeightType,
+    WeightFormat,
+    device_version,
+    df,
+    activation_potentiation
+    )
+
+    return model
+
+
+
+def predict_from_storage(Input, model_name, cuda=False, model_path=''):
+
+    """
+    Function to make a prediction
+    from storage
+
+    Args:
+        Input (list or ndarray): Input data for the model.
+        
+        model_name (str): Name of the model.
+        
+        cuda (bool, optional): CUDA GPU acceleration ? Default = False.
+        
+        model_path (str): Path of the model. Default: ''
+    
+    Returns:
+        ndarray: Output from the model.
+    """
+    
+    if cuda:
+        if not isinstance(Input, cp.ndarray): Input = cp.array(Input)
+        from .cuda.activation_functions import apply_activation
+        from .cuda.data_ops import standard_scaler
+    
+    else:
+        from .cpu.activation_functions import apply_activation
+        from .cpu.data_ops import standard_scaler
+
+    try:
+
+        model = load_model(model_name, model_path)
+        
+        model_type = model.model_type
+        activations = model.activations
+        activation_potentiation = model.activation_potentiation
+        scaler_params = model.scaler_params
+        W = model.weights
+
+        if cuda and scaler_params is not None:
+            if not isinstance(scaler_params[0], cp.ndarray): scaler_params[0] = cp.array(scaler_params[0])
+            if not isinstance(scaler_params[1], cp.ndarray): scaler_params[1] = cp.array(scaler_params[1])
+
+        Input = standard_scaler(None, Input, scaler_params)
+
+        if isinstance(activations, str):
+            activations = [activations]
+        elif isinstance(activations, list):
+            activations = [item if isinstance(item, list) or isinstance(item, str) else [item] for item in activations]
+
+        if model_type == 'MLP':
+            layer = Input
+            for i in range(len(W)):
+                if i != len(W) - 1 and i != 0: layer = apply_activation(layer, activations[i])
+
+                layer = layer @ cp.array(W[i]).T if cuda else layer @ W[i].T
+
+            result = layer
+
+        if model_type == 'PLAN':
+
+            Input = apply_activation(Input, activations)
+            result = Input @ cp.array(W).T if cuda else Input @ W.T
+        
+        if model_type == 'PTNN':
+
+            if isinstance(activation_potentiation, str):
+                activation_potentiation = [activation_potentiation]
+            elif isinstance(activation_potentiation, list):
+                activation_potentiation = [item if isinstance(item, list) or isinstance(item, str) else [item] for item in activation_potentiation]
+
+            Input = apply_activation(Input, activation_potentiation)
+            layer = Input @ cp.array(W[0]).T if cuda else Input @ W[0].T
+
+            for i in range(1, len(W)):
+                if i != len(W) - 1: layer = apply_activation(layer, activations[i])
+
+                layer = layer @ cp.array(W[i]).T if cuda else layer @ W[i].T
+
+            result = layer
+
+        return result
+    
+    except Exception as e:
+        raise RuntimeError(Fore.RED + f"ERROR: An error occurred in predict_from_storage {e}" + Style.RESET_ALL) from e
+
+
+def reverse_predict_from_storage(output, model_name, cuda=False, model_path=''):
+
+    """
+    reverse prediction function from storage
+    Args:
+
+        output (list or ndarray): output layer for the model .
+
+        model_name (str): Name of the model.
+
+        cuda (bool, optional): CUDA GPU acceleration ? Default = False.
+        
+        model_path (str): Path of the model. Default: ''
+
+    Returns:
+        ndarray: Input from the model.
+    """
+
+    if cuda:
+        if not isinstance(output, cp.ndarray): output = cp.array(output)
+
+    model = load_model(model_name, model_path)
+    
+    W = model.weights if not cuda else cp.array(model.weights)
+
+    try:
+        Input = W.T @ output
+        return Input
+    except Exception as e:
+        raise RuntimeError(Fore.RED + f"ERROR: An error occurred {e}" + Style.RESET_ALL) from e
+    
+
+
+def predict_from_memory(Input, model, cuda=False):
+
+    """
+    Function to make a prediction.
+    from memory.
+
+    Args:
+        Input (list or ndarray): Input data for the model.
+
+        model (tuple): Trained model.
+
+        cuda (bool, optional): CUDA GPU acceleration ? Default = False.
+        
+    Returns:
+    ndarray: Output from the model.
+    """
+    
+    model_type = model.model_type
+    activations = model.activations
+    activation_potentiation = model.activation_potentiation
+    scaler_params = model.scaler_params
+    W = model.weights
+
+    if not cuda: 
+        from .cpu.data_ops import standard_scaler
+        from .cpu.activation_functions import apply_activation
+    else:
+        if scaler_params is not None:
+            if not isinstance(scaler_params[0], cp.ndarray): scaler_params[0] = cp.array(scaler_params[0])
+            if not isinstance(scaler_params[1], cp.ndarray): scaler_params[1] = cp.array(scaler_params[1])
+
+        if not isinstance(Input, cp.ndarray): Input = cp.array(Input)
+        from .cuda.data_ops import standard_scaler
+        from .cuda.activation_functions import apply_activation
+
+    if model_type != 'PLAN' and model_type != 'MLP' and model_type != 'PTNN': raise ValueError("model_type parameter must be 'PLAN', 'MLP' or 'PTNN'.")
+
+    try:
+    
+        Input = standard_scaler(None, Input, scaler_params)
+        
+        if isinstance(activations, str):
+            activations = [activations]
+        elif isinstance(activations, list):
+            activations = [item if isinstance(item, list) or isinstance(item, str) else [item] for item in activations]
+
+        if model_type == 'MLP':
+            layer = Input
+            for i in range(len(W)):
+                if i != len(W) - 1 and i != 0: layer = apply_activation(layer, activations[i])
+
+                layer = layer @ cp.array(W[i]).T if cuda else layer @ W[i].T
+
+            result = layer
+
+        if model_type == 'PLAN':
+
+            Input = apply_activation(Input, activations)
+            result = Input @ cp.array(W).T if cuda else Input @ W.T
+        
+        if model_type == 'PTNN':
+
+            if isinstance(activation_potentiation, str):
+                activation_potentiation = [activation_potentiation]
+            elif isinstance(activation_potentiation, list):
+                activation_potentiation = [item if isinstance(item, list) or isinstance(item, str) else [item] for item in activation_potentiation]
+
+            Input = apply_activation(Input, activation_potentiation)
+            layer = Input @ cp.array(W[0]).T if cuda else Input @ W[0].T
+
+            for i in range(1, len(W)):
+                if i != len(W) - 1: layer = apply_activation(layer, activations[i])
+
+                layer = layer @ cp.array(W[i]).T if cuda else layer @ W[i].T
+
+            result = layer
+        
+        return result
+
+    except Exception as e:
+        raise RuntimeError(Fore.RED + f"ERROR: An error occurred in predict_from_memory {e}" + Style.RESET_ALL) from e
+
+def reverse_predict_from_memory(output, W, cuda=False):
+
+    """
+    reverse prediction function from memory
+
+    Args:
+
+        output (list or ndarray): output layer for the model.
+
+        W (ndarray): Weights of the model.
+
+        cuda (bool, optional): CUDA GPU acceleration ? Default = False.
+
+    Returns:
+        ndarray: Input from the model.
+    """
+
+    try:
+        if cuda: W = cp.array(W)
+        Input = W.T @ output
+        return Input
+    
+    except Exception as e:
+        raise RuntimeError(Fore.RED + f"ERROR: An error occurred {e}" + Style.RESET_ALL) from e
+
+
+def get_weights():
+
+    return 0
+
+
+def get_preds():
+
+    return 1
+
+
+def get_acc():
+
+    return 2
+
+
+def get_act():
+
+    return 3
+
+
+def get_scaler():
+
+    return 4
+
+
+def get_preds_softmax():
+
+    return 5
+
+
+def get_model_type():
+
+    return 6
+
+
+def get_weights_type():
+     
+    return 7
+
+
+def get_weights_format():
+     
+    return 8
+
+
+def get_model_version():
+     
+    return 9
+
+
+def get_model_df():
+     
+    return 10
+
+
+def get_act_pot():
+     
+    return 11