ValueError：无法将字符串转换为浮点数：'horse'

Question

I am working one Keras CIFAR10 Learning experiment, and the images got them from Kaggle, which is a CSV file with two columns, on 'id', the other 'label'. from here I do this. I know that I need to convert my labels to tensors, but don't know how to do it. I looked in the internet everywhere of how, but couldn't find anything that deals with reading CSV file from kaggle. Maybe this is not the way to do this....

here is the link https://www.kaggle.com/c/cifar-10 but there are not kernels as an example.

Thanks in advance for your help.

I am using from tensorflow.keras.xxxxxx

import pandas as pd
print("Image IDs and Labels (TRAIN)")
train_df = pd.read_csv(TRAIN_DF_PATH)

# Add extension to id_code to train images
train_df['id'] = train_df['id'].apply(str) + ".png"

display(train_df.head())

def preprocess_image(path, sigmaX=40):
    """
    The whole preprocessing pipeline:
    1. Read in image
    3. Resize image to desired size
    """
    image = cv2.imread(path)
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    image = cv2.resize(image, (IMG_WIDTH, IMG_HEIGHT))

    return image

# Add Image augmentation to our generator
train_datagen = ImageDataGenerator(rotation_range=360,
                                   horizontal_flip=True,
                                   vertical_flip=True,
                                   validation_split=0.25,
                                   rescale=1. / 255)

# Use the dataframe to define train and validation generators
train_generator = train_datagen.flow_from_dataframe(train_df, 
                                                    x_col='id', 
                                                    y_col='label',
                                                    directory = TRAIN_IMG_PATH,
                                                    target_size=(IMG_WIDTH, IMG_HEIGHT),
                                                    batch_size=BATCH_SIZE,
                                                    class_mode='other',
                                                    preprocessing_function=preprocess_image, 
                                                    subset='training')

val_generator = train_datagen.flow_from_dataframe(train_df, 
                                                  x_col='id', 
                                                  y_col='label',
                                                  directory = TRAIN_IMG_PATH,
                                                  target_size=(IMG_WIDTH, IMG_HEIGHT),
                                                  batch_size=BATCH_SIZE,
                                                  class_mode='other',
                                                  preprocessing_function=preprocess_image, 
                                                  subset='validation')

Batch_Size  = 64
epochs      = 25

# loop over the number of models to train
for i in np.arange(0, 5):

    # initialize the optimizer and model
    print("[INFO] training model {}/{}".format(i + 1, 5))
    opt = Adam(lr=1e-5)

    conv_base = ResNet50(weights='imagenet', include_top=False, input_shape=(32, 32, 3))

    model = models.Sequential()
    model.add(conv_base)
    model.add(layers.UpSampling2D((2,2)))
    model.add(layers.UpSampling2D((2,2)))
    model.add(layers.UpSampling2D((2,2)))
    model.add(layers.Flatten())
    model.add(layers.BatchNormalization())
    model.add(layers.Dense(128, activation='relu'))
    model.add(layers.Dropout(0.5))
    model.add(layers.BatchNormalization())
    model.add(layers.Dense(64, activation='relu'))
    model.add(layers.Dropout(0.5))
    model.add(layers.BatchNormalization())
    model.add(layers.Dense(10, activation='softmax'))

    early_stop = EarlyStopping('val_loss', patience=5)
    reduce_lr = ReduceLROnPlateau('val_loss', factor=0.01, patience=3, verbose=1)

    ############################################################################
    trained_models_path = './best_model_adam/'
    model_names = trained_models_path + 'epoch_{epoch:02d}_val_acc_{val_acc:.4f}_'
    model_checkpoint = ModelCheckpoint(model_names +"model_{}.hdf5".format(i), verbose=1, save_best_only=True)
    ############################################################################

    callbacks = [model_checkpoint, early_stop, reduce_lr]

    #model.compile(optimizer=optimizers.RMSprop(lr=2e-5), loss='binary_crossentropy', metrics=['acc'])
    model.compile(optimizer=Adam(lr=1e-5), loss='binary_crossentropy', metrics=['acc'])

    # train the network
    history = model.fit_generator(
                            train_generator,
                            epochs = epochs,
                            steps_per_epoch= train_df.shape[0] // Batch_Size,
                            validation_data= val_generator,
                            validation_steps = val_generator.shape[0] // Batch_Size,
                            #batch_size = Batch_Size, 
                            verbose=1,
                            callbacks = [model_checkpoint, early_stop]
                        )

    # save the model to disk
    p = ["./models/model_{}.model".format(i)]
    model.save(os.path.sep.join(p))

    # evaluate the network
    predictions = model.predict(testX, batch_size=64)
    report = classification_report(testY.argmax(axis=1), predictions.argmax(axis=1), target_names=labelNames)

    # save the classification report to file
    p = ["./output/model_{}.txt".format(i)]
    f = open(os.path.sep.join(p), "w")
    f.write(report)
    f.close()

When I run the fit_generator I get his error

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/constant_op.py in constant(value, dtype, shape, name)
    244   """
    245   return _constant_impl(value, dtype, shape, name, verify_shape=False,
--> 246                         allow_broadcast=True)
    247 
    248 

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/constant_op.py in _constant_impl(value, dtype, shape, name, verify_shape, allow_broadcast)
    252   ctx = context.context()
    253   if ctx.executing_eagerly():
--> 254     t = convert_to_eager_tensor(value, ctx, dtype)
    255     if shape is None:
    256       return t

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/constant_op.py in convert_to_eager_tensor(value, ctx, dtype)
    113     return t
    114   else:
--> 115     return ops.EagerTensor(value, handle, device, dtype)
    116 
    117 

ValueError: could not convert string to float: 'horse'

Answer 1

You can convert your category labels to numbers and then make a new column for those numbers. scikit-learn has a builtin for this but its easy enough without that:

import pandas as pd
df = pd.DataFrame({'label':['cat','dog','horse'],'b':[1,2,3]})
all_labels= df.label.unique().tolist()
all_labels.sort()
label_to_number={label:all_labels.index(label) for label in all_labels}
df['label_num']=df.apply(lambda r:label_to_number[r.label],axis=1)

Now you can send label_number to your training (y_col='label_number') . This is all assuming that integer cateories are ok and you don't need 'one-hot encoding' - if you do then again scikit has provision for that. From here it seems like the integer categories are fine however.

Answer 2

@jeremy_rutman, Thanks! I got it working

import pandas as pd
print("Image IDs and Labels (TRAIN)")
train_df = pd.read_csv(TRAIN_DF_PATH)

# Add extension to id_code to train images
train_df['id'] = train_df['id'].apply(str) + ".png"

all_labels = train_df['label'].unique().tolist() 
all_labels.sort() 
label_to_number={label:all_labels.index(label) for label in all_labels} 
train_df['label']=train_df.apply(lambda r:label_to_number[r.label],axis=1)


display(train_df.head())
print(train_df['id'])
```
The model if fitting now, but for some reason, my two GPU's cards 
are not kicking in.... I think lots of things got broken with TensorFlow
 2.0, but that is another topic...
thanks a lot for your help.