I am working on a movie genre prediction using poster images. In which I have created a model pipeline where I am giving image path as an input and preprocessing it in the pipeline and at last, it is giving me the prediction(movie genre). But I want to change the format of the output, I have tried creating the predict function but it's not working.
In this transformer, I am converting my image path into the NumPy array
from sklearn.base import BaseEstimator, TransformerMixin
class RGB2GrayTransformer(BaseEstimator, TransformerMixin):
"""
Convert an array of RGB images to grayscale
"""
def __init__(self,datafile='/content/Movies-Poster_Dataset/train.csv'):
self.df = pd.read_csv(datafile)
self.df = self.df.iloc[:1000,:]
def fit(self, X, y=None):
"""returns itself"""
return self
def transform(self, X, y=None):
"""perform the transformation and return an array"""
l = np.empty(shape=[850,350,350,3])
if isinstance(X, str):
X = image.load_img(X, target_size=(img_width, img_height, 3))
img = image.img_to_array(X)
img = img/255.0
img = img.reshape(1, img_width, img_height, 3)
np.append(l,img)
else:
for img in X:
img = image.img_to_array(img)
img = img/255.0
img = img.reshape(1, img_width, img_height, 3)
np.append(l,img)
return l
In the below, I am creating a Keras model and predicting my result
def get_training_model():
input_layer = tf.keras.layers.Input(shape=(350,350,3),name="input_layer")
c1=Conv2D(16, (3,3), activation='relu', input_shape = X_train[0].shape)(input_layer)
c3=BatchNormalization()(c1)
c4=MaxPool2D(2,2)(c3)
c5=Dropout(0.3)(c4)
c6=Conv2D(32, (3,3), activation='relu')(c5)
c7=BatchNormalization()(c6)
c8=MaxPool2D(2,2)(c7)
c9=Dropout(0.3)(c8)
c10=Conv2D(64, (3,3), activation='relu')(c9)
c11=BatchNormalization()(c10)
c12=MaxPool2D(2,2)(c11)
c13=Dropout(0.4)(c12)
c14=Conv2D(128, (3,3), activation='relu')(c13)
c15=BatchNormalization()(c14)
c16=MaxPool2D(2,2)(c15)
c17=Dropout(0.5)(c16)
c18=Flatten()(c17)
c19=Dense(128, activation='relu')(c18)
c20=BatchNormalization()(c19)
c21=Dropout(0.5)(c20)
c23=Dense(128, activation='relu')(c21)
c24=BatchNormalization()(c23)
c25=Dropout(0.5)(c24)
outputs=Dense(25, activation='sigmoid')(c25)
# Create the model
model = tf.keras.models.Model(input_layer, outputs)
# Compile the model and return it
model.compile(optimizer='adam', loss = 'binary_crossentropy', metrics=['accuracy'])
return model
And then my pipeline is looking something like this
HOG_pipeline = Pipeline([
('grayify', RGB2GrayTransformer()),
('final',get_training_model())
])
and the output I am getting :
whereas the output I want :
Using the below code I can get my output in the desired format but I don't know how to fit this code in my pipeline
top3 = np.argsort(y_prob[0])[:-4:-1]
l=[]
for i in range(3):
l.append(classes[top3[i]])
print(l)