logo

PYTORCH-CLASSIFIER

Gallery details

This project builds a deep learning network to identify 102 different types of flowers. The dataset was obtained from the 102 category flowers dataset. While this specific example is used on this data, this model can be trained on any set of labeled images. Below are a few examples of the variability between classes and within the classes themselves.

CLASS VARIABILITY BETWEEN CLASSES-

The 3 images below are: (Spear Thistle) (Fire Lily) (Cantenbury Bells)

CLASS VARIABILITY WITHIN CLASSES-

Each of the 3 images below is a Toad Lily

ARCHITECTURE-

Click for Github – Project Source Code

Deep Learning Network Testing Snippets

1.DEVELOPING THE APPLICATION –

The project is broken down into multiple steps:

  • Load and preprocess the image dataset
  • Train the image classifier on the dataset
  • Use the trained classifier to predict image content
Data | Network | Utility Classes & Functions
Data LoadNetwork ClassNet UtilityLoad NetworkImage ProcessData MappingPredictPlot Results
# Imports here
import torch
import torchvision.transforms as tf
import torchvision.datasets as ds
import torchvision.models as models
from torch import nn
import matplotlib.pyplot as plt
import numpy as np
from PIL import Image

from neural_net import Neural_Network as net


# TODO: Define your transforms for the training, validation, and testing sets
# TODO: Load the datasets with ImageFolder
# TODO: Using the image datasets and the trainforms, define the dataloaders

def generate_datasets(params_dict, types_list, resize = 300, crop_size = 224):

    ''' Generators and data manipulation. Generates the required data transformations 
    for us to train properly. 

    Args:
        params_dict (dict): The nested dictionary containing the 'dir', 'batch' and 'shuffle' data.
        types_list (list of str): The list of param_dict keys, 'train', 'validate', 'test'.
        resize (int): The value to resize the image to.
        crop_size (int): The value we want to crop the image to

    Raises:
        TODO: Add exceptions

    Returns:
        datasets, dataloaders (tuple): The datasets and data loaders 
    '''

    # Define the transforms
    transforms = {}
    for t in types_list:

        transform_list = []
        transform_list.append(tf.Resize(resize))
        transform_list.append(tf.CenterCrop(crop_size))
        transform_list.append(tf.ToTensor())
        transform_list.append(tf.Normalize(mean=[0.485, 0.456, 0.406],std=[0.229, 0.224, 0.225]))

        if t == 'train':
            transform_list.pop(1)
            transform_list.insert(1, tf.RandomResizedCrop(crop_size))
            transform_list.insert(2, tf.RandomHorizontalFlip())

        transforms[t] = tf.Compose(transform_list)

    # Load the data sets, use dict comprehension to generate key vals for each type
    datasets = {i: ds.ImageFolder(params_dict[i]['dir'], 
                                transforms[i]) for i in types_list}
    # Define the loaders using the datasets and the transforms
    dataloaders = {i: torch.utils.data.DataLoader(datasets[i], 
                                                params_dict[i]['batch'], 
                                                params_dict[i]['shuffle']) 
                                                for i in types_list}

    return datasets, dataloaders
    

data_dir = 'flowers'
train_dir = data_dir + '/train'
valid_dir = data_dir + '/valid'
test_dir = data_dir + '/test'


# generate datasets and loaders
params_dict = {'train': {'dir': train_dir, 'batch': 64, 'shuffle': True},
               'validate':{'dir': valid_dir, 'batch': 64, 'shuffle': True},
               'test':{'dir': test_dir, 'batch': 64, 'shuffle': False}}

datasets, dataloaders = generate_datasets(params_dict, list(params_dict.keys()))
# build network

class Neural_Network(nn.Module):
    '''
    The neural network object sits a level above the classifier to
    store relevant properties and values. The classifier uses nn.LogSoftmax so use the 
    negative log likelihood loss criterion nn.NLLLoss

    Args:
        inputs (int): The number of inputs.
        hidden_sizes (list of ints): The hidden layer sizes.
        outputs (int): The number of outputs.
        hidden_activation (str): The hidden layer activation functions (ex. relu, sigmoid, tahn).
        device (str): The gpu or the cpu.
        optimizer_name (str): The optimizer name ('sgd' or 'adam') to update the weights and gradients
        dropout (float): The dropout rate, value to randomly drop input units through training.
        learn_rate (float): The learning rate value, used along with the gradient to update the weights,
            small values ensure that the weight update steps are small enough.

    Attributes:
        inputs (int): This is where we store the input count,
        hidden_sizes (list of int): This is where we store the hidden layer sizes,
        outputs (int): This is where we store the output size,
        hidden_activation (str): This is where we store the hidden activation type,
        dropout (float): This is where we store the random input unit dropout rate,
        learn_rate (float): This is where we store the learn rate value,
        processing_device (str): This is where we store the device to calculate the results,
        linear_layers (list): This is where we store the values to sequentially build the classifier,
        model (torch.nn.module or torchvision model): Where either the generated classifier or the loaded model is stored,
        optimizer (torch.optim): This is where we store the optimizer used,
        criterior (torch.nn.module.loss): This is where we store the loss function type,
        device (str): This is where we store the device,
        epochs_completed (int): This is where we store how many total epochs of training this model has.

    '''

    def __init__(self, inputs, hidden_sizes, 
                 outputs, hidden_activation, device,
                 dropout = 0.3, learn_rate = 0.002):

        super().__init__()
        # Props
        self.inputs = inputs
        self.hidden_sizes = hidden_sizes
        self.outputs = outputs
        self.hidden_activation = hidden_activation
        self.dropout = dropout
        self.learn_rate = learn_rate
        self.processing_device = device
        # Layers
        self.linear_layers = []
        self.data = hidden_sizes
        self.data.insert(0,inputs)
        self.data.append(outputs)
        # Model Stuff
        self.model, self.optimizer =  None, None
        self.criterion = nn.NLLLoss()
        self.device = device

        self.epochs_completed = 0

        self.generate_classifier()

    def generate_classifier(self):
        '''Generates the nn.module container Sequential classfier as the default for this class.

        Args:
            None.

        Raises:
            TODO: Update exceptions with error_handling class.

        Returns:
            None.
        '''

        self.linear_layers = []
        n = len(self.data)
        for i in range(n-1):
            
            self.linear_layers.append(nn.Linear(self.data[i],self.data[(i + 1) % n]))
            
            if i != n-2:
                if self.hidden_activation == 'relu':
                    self.linear_layers.append(nn.ReLU())
                elif self.hidden_activation == 'sigmoid':
                    self.linear_layers.append(nn.Sigmoid())
                elif self.hidden_activation == 'tanh':
                    self.linear_layers.append(nn.Tanh())
                self.linear_layers.append(nn.Dropout(self.dropout))

        self.linear_layers.append(nn.LogSoftmax(dim = 1))
        # expand the list into sequential args
        self.model = nn.Sequential(*self.linear_layers)

    def train_network(self, train_data, validation_data, epochs = 1, load_best_params = False, plot = False):
        '''Trains the model, requires the criterion and optimizer to be passed into the class args before hand.

        TODO: add exception handling for optimizer and criterion as None values.

        Args:
            train_data (torch.utils.data.dataloader.DataLoader): The training torch data loader.
            validation_data (torch.utils.data.dataloader.DataLoader): The validation torch data loader.
            epochs (int): The number of epochs for training.
            load_best_params (bool): If true then we will load the model_state_dict from the highest accuracy iteration
            plot (bool): If true we plot both losses.

        Raises:
            TODO: Add exceptions.

        Returns:
            None.
        '''        
        
        # move the model to whatever device we have
        self.model.to(self.device)

        # if we loaded the model in eval mode and want to train switch it
        if not self.model.training:
            self.model.train()

        iteration, running_loss = 0, 0
        highest_accuracy, high_acc_iter, high_acc_epoch = 0, 0, 0
        training_loss_set, validation_loss_set = [], []
        best_params = None

        for epoch in range(epochs):
            batch_iteration = 0
            for x, y_labels in train_data:
                # move to whatever device we have
                x, y_labels = x.to(self.device), y_labels.to(self.device)
                # zero out the gradients
                self.optimizer.zero_grad()
                # forward pass - get the log probabilities (logits / scores)
                output = self.model(x)
                # calculate the loss
                loss = self.criterion(output, y_labels)
                # backprop - calculate the gradients for the parameters
                loss.backward()
                # parameter update based on gradient
                self.optimizer.step()
                # update stats
                running_loss += loss.item()
                iteration += 1
                batch_iteration += 1

            else:
                # Validation Process
                validation_loss, accuracy = self.validate_network(validation_data)

                training_loss = running_loss/len(train_data)
                print('Model has a total of {} training epochs completed.'.format(self.epochs_completed))
                print('Active session Epoch {} out of {}'.format(epoch + 1, epochs))
                print('Currently model has Accuracy of {}% \nCurrent training loss is {} \
                    \nCurrent validation loss is {}'.format(accuracy, 
                    training_loss, validation_loss))

                training_loss_set.append(training_loss)
                validation_loss_set.append(validation_loss)
                
                print('-------------')
                running_loss = 0

                # Track best run
                if accuracy > highest_accuracy:
                    highest_accuracy = accuracy
                    high_acc_iter = batch_iteration
                    high_acc_epoch = epoch + 1
                    if load_best_params:
                        best_params = copy.deepcopy(self.model.state_dict())

                # Set the model back to train mode, enable dropout again
                self.model.train()
            self.epochs_completed += 1

        t_slope, v_slope = self.check_overfitting(training_loss_set, validation_loss_set, plot)
        print('Slope of linear reg training curve fit is {} \nSlope of linear reg Validation curve fit is {}'.format(t_slope, 
                                                                                                                    v_slope))
        print('Training session highest accuracy was {} on epoch {} batch iteration {}'.format(highest_accuracy, 
                                                                                             high_acc_epoch, 
                                                                                             high_acc_iter))
        if load_best_params:
            self.model.load_state_dict(best_params)
            print('Params from {} epoch, {} batch iteration were loaded'.format(high_acc_epoch, high_acc_iter))

    def validate_network(self, data):
        '''Validate our model to check the loss and accuracy.

        Args:
            data (torch.utils.data.dataloader.DataLoader): The data we want to validate as torch data loader.

        Raises:
            TODO: Add exceptions.

        Returns:
            loss,accuracy (tuple): The loss and accuracy of the validation.
        '''
        
        # enable eval mode, turn off dropout
        self.model.eval()
        # turn off the gradients since we are not updating params
        with torch.no_grad():
            batch_loss = 0
            batch_accuracy = 0
            # validation pass
            for x, y_labels in data:
                # move to device
                x, y_labels = x.to(self.device), y_labels.to(self.device)
                output = self.model(x)
                # update loss and extract tensor as python float
                batch_loss += self.criterion(output, y_labels).item()
                # calculate the probability
                probability = torch.exp(output)
                # get the top n indexes and values
                _, top_class = probability.topk(1, dim=1)
                # reshape top class to match label and get binary value from equals, 
                # check if the prediction matches label
                equals = top_class == y_labels.view(*top_class.shape)
                # have to convert byte tensor to float tensor and get accuracy
                batch_accuracy += torch.mean(equals.type(torch.FloatTensor)).item()

            test_accuracy = (batch_accuracy / len(data))*100
            test_loss = batch_loss / len(data)

            return test_loss, test_accuracy
    
    def check_overfitting(self, train_losses, validation_losses, plot = False):
        '''Validate our model to check the loss and accuracy

        Args:
            train_losses (list of floats): The list of training losses per epoch.
            validation_losses (list of floats): The list of validation losses per epoch.
            plot (bool): If true we plot both losses.

        Raises:
            TODO: Add exceptions.

        Returns:
            slopes (tuple): The slopes of the linear reg curve fits for both validation/training.
        '''
        # Data 
        tl_x_val = np.arange(0, len(train_losses))
        vl_x_val = np.arange(0, len(validation_losses))   
        # To numpy
        train_data = np.array([tl_x_val, train_losses])
        validate_data = np.array([vl_x_val, validation_losses])
        # Least squares polynomial fit.
        train_slope, train_intercept = np.polyfit(train_data[0], train_data[1], 1)
        validation_slope, validation_intercept = np.polyfit(validate_data[0], validate_data[1], 1)

        if plot:
            plt.plot(train_data[0], train_data[1], 'o', label='training loss')
            plt.plot(validate_data[0], validate_data[1], 'o', label='validation loss')
            plt.plot(train_data[0], train_intercept + train_slope*train_data[0], 'r', label='train_regg')
            plt.plot(validate_data[0], validation_intercept + validation_slope*validate_data[0], 'r', label='val_regg')
            plt.legend()
            plt.show()
        
        return train_slope, validation_slope

    def save_model_checkpoint(self, full_path, training_class_to_idx):
        '''Save the model checkpoint.

        Args:
            full_path (str): The full path to save the checkpoint to
            training_class_to_idx (dic of ints): This is where we store the dictionary mapping the name of the class to the index (label)

        Raises:
            TODO: Add exceptions

        Returns:
            None
        '''
                
        net_data_dic = {'input_count': self.inputs,
                        'hidden_sizes': self.hidden_sizes,
                        'outputs': self.outputs,
                        'h_activation': self.hidden_activation,
                        'dropout': self.dropout,
                        'learn_rate': self.learn_rate,
                        'epochs_completed' : self.epochs_completed}
        
        checkpoint = {'data' : net_data_dic,
                      'model' : self.model, 
                      'classifier' : self.model.classifier,
                      'optimizer.state_dict' : self.optimizer.state_dict(),
                      'state_dict' : self.model.state_dict(),
                      'device' : self.device,
                      'class_to_idx': training_class_to_idx}

        torch.save (checkpoint, full_path)
# Net Utilities to build from tv

def net_from_torchvision(hidden_sizes, outputs, hidden_activation, device, 
                        optimizer_name = 'adam', dropout = 0.3, learn_rate = 0.002, 
                        name = 'vgg16', trained = True):

    '''
    Generates a model from torchvision, and instatiates a new Neural_Network instance which sets new model 
    as the active model. A new optimizer and criterion are also generated and assigned to the class properties.

    Args:
        hidden_sizes (list of ints): The hidden layer sizes.
        outputs (int): The number of outputs.
        hidden_activation (str): The hidden layer activation functions (ex. relu, sigmoid, tahn).
        device (str): The gpu or the cpu.
        optimizer_name (str): The optimizer name ('sgd' or 'adam') to update the weights and gradients
        dropout (float): The dropout rate, value to randomly drop input units through training.
        learn_rate (float): The learning rate value, used along with the gradient to update the weights,
            small values ensure that the weight update steps are small enough.
        name (str): The pretrained model name ('vgg16', 'resnet50', 'densenet121').
        trained (bool): If the model has been trained.

    Raises:
        TODO: Update exceptions with error_handling class.

    Returns:
        net (nn_model.Neural_Network): An instance of the Neural_Network class with the trained model
            as its model and parameters.
    '''

    model = get_pretrained_model(name, trained)
    feature_count = model.classifier[0].in_features

    net = Neural_Network(feature_count, hidden_sizes, outputs, 
                        hidden_activation, device, dropout, learn_rate)

    model.classifier = net.model
    net.model = model

    if optimizer_name != 'adam' and optimizer_name != 'sgd':
        raise ValueError('Please use either SDG or Adam as optimizers')
    elif optimizer_name == 'adam':
        net.optimizer = torch.optim.Adam(net.model.classifier.parameters(), learn_rate)
    else:
        net.optimizer = torch.optim.SDG(net.model.classifier.parameters(), learn_rate)

    net.criterion = nn.NLLLoss()

    return net  

def get_pretrained_model(name = 'vgg16', trained = True):
    '''Generates the nn.module container Sequential classfier as the default for this class.

    Args:
        name (str): The pretrained model name ('vgg16', 'resnet50', 'densenet121').
        trained (bool): If the model has been trained.

    Raises:
        TODO: Update exceptions with error_handling class.

    Returns:
        model (torchvision.models.vgg.VGG): The torch vision model specified
    '''        

    # get model from torchvision
    if name == 'vgg16':
        model = models.vgg16(pretrained = trained)
    elif name == 'resnet50':
        model = models.resnet50(pretrained = trained)
    elif name == 'densenet121':
        model = models.densenet121(pretrained = trained)
    else:
        raise ValueError('Please select from either vgg16, resnet50 or \
                        densenet121 pre-trained models')

    # freeze parameters
    for parameter in model.parameters():
        parameter.requires_grad = False

    return model
# TODO: Write a function that loads a checkpoint and rebuilds the model

def load_neural_net(filepath, mode = 'train'):
    '''
    Generates a model from torchvision, and instatiates a new Neural_Network instance which sets new model 
    as the active model. A new optimizer and criterion are also generated and assigned to the class properties.

    Args:
        file_path (str): The full path to the checkpoint
        mode (str): Mode to set the model to ('train', 'eval')

    Raises:
        TODO: Update exceptions with error_handling class.

    Returns:
        net (nn_model.Neural_Network): An instance of the Neural_Network class with the loeaded model
            as its model, parameters, criterion and optimizer.
    '''        

    print('loading_net')
    #TODO: Path validation
    checkpoint = torch.load(filepath)
    # Set Params
    inputs = checkpoint['data']['input_count']
    hidden_layers = checkpoint['data']['hidden_sizes']
    outputs = checkpoint['data']['outputs']
    activation = checkpoint['data']['h_activation']
    dropout = checkpoint['data']['dropout']
    learn_rate = checkpoint['data']['learn_rate']
    device = checkpoint['device']
    model = checkpoint['model']
    model.load_state_dict(checkpoint['state_dict'])
    # Make Network
    net = Neural_Network(inputs, hidden_layers, outputs, activation, device, dropout, learn_rate)
    net.model = model
    net.epochs_completed = checkpoint['data']['epochs_completed']

    if mode == 'train':
        net.model.train()
    elif mode == 'eval':
        net.model.eval()
    else:
        raise ValueError('Error mode needs to be either train or eval')

    net.model.classifier.class_to_idx = checkpoint['class_to_idx']
    optimizer = torch.optim.Adam(net.model.classifier.parameters(), learn_rate)
    optimizer.load_state_dict(checkpoint['optimizer.state_dict'])
    criterion = nn.NLLLoss()
    net.optimizer = optimizer
    net.criterion = criterion
    # Move to processing device
    net.model.to(device)

    return net
    

# load the model
loaded_net = load_neural_net('checkpoint_1.pth', 'eval')
# process image

def process_image(image, width, height):
    ''' Scales, crops, and normalizes a PIL image for a PyTorch model,
        returns an torchNumpy array

    Args:
        image (nn_model.Neural_Network): The Neural_Network instance to use for the prediction.
        width (int): The path to the image we want to test
        height (int): The label map with the class names

    Raises:
        TODO: Add exceptions

    Returns:
        t_image (torch.Tensor): 
    '''

    # open and resize
    img = Image.open(image)
    img = img.resize((width,height))

    # crop
    current_width, current_height = img.size
    left = (current_width - width)/2
    top = (current_height - height)/2
    right = left + width
    bottom = top + height
    img = img.crop((left, top, right, bottom))

    # normalize the values
    mean = np.array([0.485, 0.456, 0.406])
    std = np.array([0.229, 0.224, 0.225])
    np_img = np.array(img) / 255
    np_img = (np_img - mean) / std

    # swap color channel position
    np_img = np.transpose(np_img, (2,0,1))

    # conver to tensor from numpy ndarray
    t_image = torch.from_numpy(np_img)

    return t_image
def map_idx_to_classes(class_to_idx, classes, predicted_idx, predicted_prob):
    '''
    Maps the predicted indexes to the keys which we need to retrieve the class names. Since our model gives us the 'value',
    we need to find the key for our class_to_idx dict, once we have the key we can use it to find the class mapping 
    (flower name in this case).

    Args:
        class_to_idx (dic of ints): This is where we store the dictionary mapping the name of the class to the index (label).
        classes (dict of strs): Dict containing the mapping of the class idx to the name.
        predicted_idx (list of ints): The topk list of predicted indexes.
        predicted_prob (list of floats): The probability list from topk.

    Raises:
        TODO: Update exceptions with error_handling class.

    Returns:
        idx_classes_dict (dict): Dictionary containing 'predicted_indexes': indexes predicted by network, 
                                                        'idx_to_class': mapped idx_to_class, 
                                                        'classes': class names,
                                                        'probabilites': the probabilities for classes.

    '''

    idx_classes_dict = {}
    predicted_class_names = []
    predicted_idx_to_class = []

    for x in predicted_idx:
        for k,v in class_to_idx.items():
            if x == v:
                predicted_class_names.append(classes[k])
                predicted_idx_to_class.append(k)

    idx_classes_dict['predicted_idx'] = predicted_idx
    idx_classes_dict['classes'] = predicted_class_names
    idx_classes_dict['idx_to_class'] = predicted_idx_to_class
    idx_classes_dict['probabilities'] = predicted_prob

    return idx_classes_dict
# helper methods

def predict(network, image_path, class_names, topk=5):
    ''' Predict the class (or classes) of an image using a trained deep learning model.

    Args:
        network (nn_model.Neural_Network): The Neural_Network instance to use for the prediction.
        image_path (str): The path to the image we want to test
        class_names (dict of ints): The label map with the class names
        topk (int): The number of top probabilities and classes we want.

    Raises:
        TODO: Add exceptions

    Returns:
        data_dict (dict): Dictionary containing 'predicted_indexes': indexes predicted by network, 
                                                'idx_to_class': mapped idx_to_class, 
                                                'classes': class names,
                                                'probabilites': the probabilities for classes.        
    '''

    # convert image 
    img = process_image(image_path, 224, 224)
    # need to pass the image tensor with first argument of n where n represents our batch size
    img.unsqueeze_(0)
    # move to device
    img.to(network.device)
    
    # generate the prediction
    network.model.to(network.device)
    # enable eval mode, turn off dropout
    network.model.eval()
    # turn off the gradients since we are not updating params
    with torch.no_grad():
        img = img.to(network.device, dtype=torch.float)
        # get the log softmax
        output = network.model(img)
        # get the prob
        probabilities = torch.exp(output)
        # get the top k values
        top_probabilities, top_classes = probabilities.topk(topk, dim=1)
        
    # remove the tensor cuda by moving to cpu, squeeze to remove dimensions and send to list to index
    top_probabilities = top_probabilities.cpu().squeeze().tolist()
    top_classes = top_classes.cpu().squeeze().tolist()
    
    # generate the idx_to_class mapping dict
    data_dict = map_idx_to_classes(network.model.classifier.class_to_idx, class_names, top_classes, top_probabilities)

    return data_dict
# TODO: Display an image along with the top 5 classes

# To visualize more than 1 result at a time I added this function, displays a grid of n results with image and prediction

import numpy as np
import seaborn as sb
import matplotlib.pyplot as plt
import pandas as pd

def plot_image_results(datasets, filter, count):
    # generate random n indexes to choose random testing images from dataset
    idx = np.random.randint(0,len(datasets[filter].imgs),size=(count,))
    print(idx)

    # get the image folder number idx from the randomly selected dataset image
    batch_idx = [datasets[filter].imgs[x][0].split('\\')[-2] for x in idx]
    print(batch_idx)

    # fix the full path for the batch idx's
    batch_paths = [datasets[filter].imgs[x][0].replace('\\','/') for x in idx]
    print(batch_paths)

    # get actual flower name from the mapping back to the label
    labeled_names = [flowers_to_name[x] for x in batch_idx]
    print(labeled_names)

    # zip the data
    data = dict(zip(labeled_names, batch_paths))

    # set the subplots
    rows = (len(data.items())) 
    cols = 2
    fig, axs = plt.subplots(nrows = rows, ncols= cols, figsize=(cols*4,rows*3), squeeze = False)
    axs = axs.flatten()
    plt.tight_layout()

    # iterate through the dict, plot the graphs on the even grid cell
    # plot matching imgs on the odd grid cells
    count, img_counter = 0, 1
    for name, path in data.items():
        # get the predictions
        results_dict = predict(loaded_net, path, flowers_to_name)
        for k,v in results_dict.items():
            print('{}:{}'.format(k, v))
        print('flower is {}\n'.format(name))
                
        # barplots for the results
        bp = sb.barplot(x=results_dict['probabilities'], y=results_dict['classes'], ax=axs[count])
        bp.set_title(name)
        
        # plot the images
        img = process_image(path, 224, 224)
        imshow(img, axs[img_counter])
         
        # increment the counters
        count += 2
        img_counter += 2
            
    plt.show()

plot_image_results(datasets, 'test', 5)

.Build and train network –

Building and training the classifier:

  • Load a pre-trained network
  • Define a new, untrained feed-forward network as a classifier, choose activation functions and dropouts
  • Train the classifier layer using backpropagation using the pre-trained network to get the features
  • Track the loss and accuracy on the validation set to determine the best hyperparameters
# TODO: Build and train your network

# check for gpu
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# network instance
neural_net = net_from_torchvision([1024,512], 102, 'relu', device, learn_rate = 0.001)

# train for 25 epochs
neural_net.train_network(dataloaders['train'], dataloaders['validate'], 5, plot = True)

 

.Test the network-

# TODO: Do validation on the test set
loss, acc = neural_net.validate_network(dataloaders['test'])
print('acc on test is {} % \nloss is {}'.format(acc, loss))

 

  • acc on test is 85.90214940217825 %
  • loss is 0.5098767114373354

.Save the checkpoint-

# TODO: Save the checkpoint 
neural_net.save_model_checkpoint('checkpoint_1.pth', datasets['train'].class_to_idx)

 

.Load the checkpoint-

# load the model
loaded_net = load_neural_net('checkpoint_1.pth', 'eval')

 

.Inference for classification-

  • probs, classes = predict(image_path, model)
  • print(probs)
  • print(classes)
  • > [ 0.01558163 0.01541934 0.01452626 0.01443549 0.01407339]
  • > [’70’, ‘3’, ’45’, ’62’, ’55’]

.Image Processing & Class Predictions

img_path = 'flowers/test/15/image_06360.jpg'

# get the predictions 
results_dict = predict(loaded_net, img_path, flowers_to_name)
for k,v in results_dict.items():
    print('{}:{}'.format(k, v))
    
""" OUTPUT
predicted_idx:[9, 62, 45, 39, 48]
classes:['yellow iris', 'black-eyed susan', 'buttercup', 'daffodil', 'common dandelion']
idx_to_class:['15', '63', '48', '42', '50']
probabilities:[0.9235654473304749, 0.04405064508318901, 0.015102370642125607, 0.010496463626623154, 0.001698569511063397]
"""
    
# get flower names fromr results dict
print(results_dict['idx_to_class'])
names = [flowers_to_name[x] for x in results_dict['idx_to_class']]
print(names)

""" OUTPUT
['15', '63', '48', '42', '50']
['yellow iris', 'black-eyed susan', 'buttercup', 'daffodil', 'common dandelion']
"""

plot_image_results(datasets, 'test', 5)

 

.Plot Results-

To visualize more than 1 result at a time I added this function, displays a grid of n results with image and prediction

 

2.COMMAND LINE APPLICATION SPECIFICATIONS –

The project submission must include at least two files train.py and predict.py. The first file, train.py, will train a new network on a dataset and save the model as a checkpoint. The second file, predict.py, uses a trained network to predict the class for an input image.

    • Train a new network on a data set with train.py :
      • Basic usage: python train.py data_directory
      • Prints out training loss, validation loss, and validation accuracy as the network trains
      • Options:
        • Set directory to save checkpoints:
          • python train.py data_dir –save_dir save_directory
        • Choose architecture: python train.py data_dir --arch "vgg13"
        • Set hyperparameters:
          • python train.py data_dir –learning_rate 0.01 –hidden_units 512 –epochs 20
        • Use GPU for training: python train.py data_dir --gpu
    • Predict flower name from an image with predict.py along with the probability of that name. That is, you’ll pass in a single image /path/to/image and return the flower name and class probability.
      • Basic usage: python predict.py /path/to/image checkpoint
      • Options:
        • Return top K most likely classes:
          • python predict.py input checkpoint –top_k 3
        • Use a mapping of categories to real names:
          • python predict.py input checkpoint –category_names cat_to_name.json
        • Use GPU for inference: python predict.py input checkpoint --gpu
Command Line Application
Predict.pyTrain.py
import argparse
import json
import torch as t
import os
import sys

from network.net_operations import Net_Operations as net_ops
from utilities.net_utils import Net_Utilities as net_utils
from utilities.utils import Utilities as utils

'''predict.py: Predict a flower name from an image along with the probability of that name '''
__author__ = "Luis Quinones"
__email__ = "luis@complicitmatter.com"
__status__ = "Prototype"

def main():

    try:

        args_dict = {}

        names = ['image_path', 'model_checkpoint_path', '--top_k', '--category_names', '--gpu']
        defaults = [None, None, 3, 'flower_to_name.json', False]
        types = [str, str, int, str, bool]
        helpers = ['the path to the image we want to predict',
                'the path to the model checkpoint to load',
                'return the top k most likely cases',
                'Json file with the mapping of categories to real names',
                'Use the gpu for computing, if no use cpu']
                
        for i in range(len(names)):
            data = {}
            data['name'] = names[i]
            data['default'] = defaults[i]
            data['type'] = types[i]
            data['help'] = helpers[i]

            args_dict[i] = data

        # get the args
        args = utils.get_input_args(args_dict)        

        # variables
        img_path = args.image_path
        model_checkpoint = args.model_checkpoint_path
        top_k = args.top_k
        categories = args.category_names
        enable_gpu = args.gpu 
        
        # check if the img path exist
        while not os.path.isfile(img_path):
            img_path = input('Image file does not exist, please input a correct path \n')
            if img_path == 'quit':
                exit()

        # check if the checkpoint file exist
        while not os.path.isfile(model_checkpoint):
            model_checkpoint = input('Model checkpoint does not exist, please input a correct path \n')
            if model_checkpoint == 'quit':
                exit()

        while top_k < 1:
            val = input('Top_k value must be greater than 0, please enter a new value \n')
            top_k = int(val)

        # check for gpu
        if not t.cuda.is_available() and enable_gpu:
            print('Your device does not have a CUDA capable device, we will use the CPU instead')
            response = input('Your device does not have a CUDA capable device, would you like to run it on the CPU instead? Enter Yes or No -> ')
            while response not in ('yes', 'no'):  
                if response.lower() == 'yes':
                    break
                elif response.lower() == "no":
                    print('exiting the program')
                    exit()
                else:
                    print('Please respond yes or no ')
            enable_gpu = False

        # load from checkpoint and set device
        mfcp = net_utils.load_neural_net(model_checkpoint, 'eval')
        mfcp.device = 'cuda' if enable_gpu else 'cpu'

        # load json data
        with open(categories, 'r') as f:
            categories_to_name = json.load(f)

        # make the predictions
        results_dict = net_ops.predict(mfcp, img_path, categories_to_name, topk = top_k)
        names = [categories_to_name[x] for x in results_dict['idx_to_class']]

        # get flower name from path
        flower_name = categories_to_name[img_path.split('/')[-2]]
        
        # print the top n results
        print('FLOWER NAME IS {} \n'.format(flower_name.upper()))
        print('THE TOP {} RESULTS ARE:'.format(top_k))
        for i, name in enumerate(names):
            print('Name = {} \nProbability = {} \n'.format(name, results_dict['probabilities'][i]))

    except Exception as ex:
        raise ex

if __name__ == "__main__":
    main()
import argparse
import sys
from collections import defaultdict
import os
import datetime
import torch as t

from utilities.utils import Utilities as utils
from utilities.net_utils import Net_Utilities as net_utils
from utilities.data_utils import Data_Utilities as data_util

'''train.py: Train a new network on a dataset of images '''
__author__ = "Luis Quinones"
__email__ = "luis@complicitmatter.com"
__status__ = "Prototype"

def main():

    try:

        args_dict = {}

        names = ['data_dir', '--save_dir', '--arch', '--learning_rate', '--hidden_units', '--epochs', '--gpu']

        def_save_path = os.getcwd() + '/checkpoint_nn_train.pth'
        def_save_path = def_save_path.replace('\\','/')
        
        defaults = [None, def_save_path, 'vgg16', 0.001, '1024, 512', 2, False]
        types = [str, str, str, float, str, int, bool]
        helpers = ['the directory of the data, ex. flowers/',
                'the fullpath with name where we want to save our checkpoints, ex. saved/checkpoint_xx.pth',
                'the architecture to transfer learning, vgg16, resnet50, densenet121',
                'the learning rate value',
                'the values for the hidden layer sizes form ex. 1024,512',
                'the number of epochs to train',
                'Use the gpu for computing, if no use cpu']
                
        for i in range(len(names)):
            data = {}
            data['name'] = names[i]
            data['default'] = defaults[i]
            data['type'] = types[i]
            data['help'] = helpers[i]

            args_dict[i] = data

        # get the args
        args = utils.get_input_args(args_dict)        

        # variables
        directory = args.data_dir
        if not os.path.exists(directory):
            raise OSError('Directory does not exist, please specify a new one')

        checkpt_dir = args.save_dir
        arch = args.arch
        learn_rate = args.learning_rate
        hidden_layers = args.hidden_units
        epochs = args.epochs 
        enable_gpu = args.gpu 

        trim = hidden_layers.strip('[]').split(',')
        hidden_layers = [int(i) for i in trim] 

        # check for gpu
        if not t.cuda.is_available() and enable_gpu:
            print('Your device does not have a CUDA capable device, we will use the CPU instead')
            response = input('Your device does not have a CUDA capable device, would you like to run it on the CPU instead? Enter Yes or No -> ')
            while response not in ('yes', 'no'):  
                if response.lower() == 'yes':
                    break
                elif response.lower() == "no":
                    print('exiting the program')
                    exit()
                else:
                    print('Please respond yes or no ')
            enable_gpu = False

        # generate datasets
        params_dict = {'train': {'dir': directory + 'train', 'batch': 64, 'shuffle': True},
                    'validate':{'dir': directory + 'valid', 'batch': 64, 'shuffle': True}}

        datasets, dataloaders = data_util.generate_datasets(params_dict, list(params_dict.keys()))

        # network instance
        processor = 'cuda' if enable_gpu else 'cpu'
        neural_net = net_utils.net_from_torchvision(hidden_layers, 102, 'relu', processor, learn_rate = learn_rate, name = arch)

        # train for n epochs
        neural_net.train_network(dataloaders['train'], dataloaders['validate'], epochs, plot = True)

        # save model
        neural_net.save_model_checkpoint(checkpt_dir, datasets['train'].class_to_idx)

    except Exception as ex:
        raise ex

if __name__ == '__main__':
    main()

.Test command line application-

  • Share

Leave a reply

Your email address will not be published.