I am writing to build a model to predict handwritten characters using the dataset given here (https://www.kaggle.com/sachinpatel21/az-handwritten-alphabets-in-csv-format)
EDIT: ( after making the changes suggested in the comments )
Error I get now : ValueError: Error when checking input: expected conv2d_4_input to have shape (28, 28, 1) but got array with shape (249542, 784, 1)
Find below the code for the CNN :
from keras.models import Sequential
from keras.layers import Dense
from keras.layers import Dropout
from keras.layers import Flatten
from keras.layers.convolutional import Conv2D
from keras.layers.convolutional import MaxPooling2D
from keras import backend as K
from keras.utils import np_utils
from sklearn.model_selection import train_test_split
import numpy as np
import pandas as pd
seed = 785
np.random.seed(seed)
dataset = np.loadtxt('../input/A_Z Handwritten Data/A_Z Handwritten Data.csv', delimiter=',')
print(dataset.shape) # (372451, 785)
X = dataset[:,1:785]
Y = dataset[:,0]
(X_train, X_test, Y_train, Y_test) = train_test_split(X, Y, test_size=0.33, random_state=seed)
X_train = X_train / 255
X_test = X_test / 255
X_train = X_train.reshape((-1, X_train.shape[0], X_train.shape[1], 1))
X_test = X_test.reshape((-1, X_test.shape[0], X_test.shape[1], 1))
print(X_train.shape) # (1, 249542, 784, 1)
Y_train = np_utils.to_categorical(Y_train)
Y_test = np_utils.to_categorical(Y_test)
print(Y_test.shape) # (122909, 26)
num_classes = Y_test.shape[1] # 26
model = Sequential()
model.add(Conv2D(32, (5, 5), input_shape=(28, 28, 1), activation='relu', data_format="channels_last"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
print("DONE")
model.fit(X_train, Y_train, validation_data=(X_test, Y_test), epochs=10, batch_size=256, verbose=2)
# Final evaluation of the model
scores = model.evaluate(X_test,Y_test, verbose=0)
print("CNN Error: %.2f%%" % (100-scores[1]*100))
model.save('weights.model')
So the problem is that your data isn't structured properly. Look at the solution below:
Read the data with pandas:
data = pd.read_csv('/users/vpolimenov/Downloads/A_Z Handwritten Data.csv')
data.shape
# shape: (372450, 785)
Get your X and y:
data.rename(columns={'0':'label'}, inplace=True)
X = data.drop('label',axis = 1)
y = data['label']
Split and scale:
X_train, X_test, y_train, y_test = train_test_split(X,y)
standard_scaler = MinMaxScaler()
standard_scaler.fit(X_train)
X_train = standard_scaler.transform(X_train)
X_test = standard_scaler.transform(X_test)
Here is the magic:
X_train = X_train.reshape(X_train.shape[0], 28, 28, 1).astype('float32')
X_test = X_test.reshape(X_test.shape[0], 28, 28, 1).astype('float32')
y_train = np_utils.to_categorical(y_train)
y_test = np_utils.to_categorical(y_test)
X_train.shape
# (279337, 28, 28, 1)
Here is your model:
num_classes = y_test.shape[1] # 26
model = Sequential()
model.add(Conv2D(32, (5, 5), input_shape=(28, 28, 1), activation='relu', data_format="channels_last"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
print("DONE")
model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=10, batch_size=256, verbose=2) # WHERE I GET THE ERROR
Summary of your model:
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_25 (Conv2D) (None, 24, 24, 32) 832
_________________________________________________________________
max_pooling2d_25 (MaxPooling (None, 12, 12, 32) 0
_________________________________________________________________
dropout_1 (Dropout) (None, 12, 12, 32) 0
_________________________________________________________________
flatten_25 (Flatten) (None, 4608) 0
_________________________________________________________________
dense_42 (Dense) (None, 128) 589952
_________________________________________________________________
dense_43 (Dense) (None, 26) 3354
=================================================================
Total params: 594,138
Trainable params: 594,138
Non-trainable params: 0
I've stopped it after the second epoch, but you can see it working:
Train on 279337 samples, validate on 93113 samples
Epoch 1/10
- 80s - loss: 0.2478 - acc: 0.9308 - val_loss: 0.1021 - val_acc: 0.9720
Epoch 2/10
- 273s - loss: 0.0890 - acc: 0.9751 - val_loss: 0.0716 - val_acc: 0.9803
Epoch 3/10
Note:
It takes so long to fit due to the huge number of parameters in your network. You can try to reduce those and get a much faster/efficient network.