Neural Network Challenge-: House price prediction

#41

Only those columns having categorical data and having nan values?

#42

df[ columns ] = df[ columns ].fillna(“unknown”) and this thing is giving me error

Screenshot (82)
Screenshot (82)1366×768 39.9 KB

#43

hey don’t copy the quotes ,
add them on your own, they are a special character here.

#44

Bhaiya i have one more doubt why my directory named as cats ,dogs,horses,humans is not making in val_images?

Screenshot (83)
Screenshot (83)1366×768 48.3 KB

#45

Screenshot (84)
Screenshot (84)1366×768 60.5 KB

#46

hey @sambher_shubham ,
just confirm that , does the path exists in your system.
The error says its incorrect.

#47

yes sir the path exists. I have make the directory val_images in my first Screen shot. After that i am making sub folders in it named as cat,dog,horses,humans and these subfolders are not making. The error is saying path is incorrect because these subfolders are not there in val_images but i have done the coding part for the subfolders directory.
You can see this

Screenshot (85)
Screenshot (85)1310×133 3.34 KB

#48

change it to…
if not os.path.exists§:
os.mkdir§

as that variable p exists but not the actual path in directory.

#49

Thank you Bhaiya. I got it

#50

How can i share my github repo so that i can show my code?

#51

just copy the link and paste it here.

#52

Thank you soo much bhaiya for all you help.

#53

:slightly_smiling_face:.

#54

https://github.com/Shubham967/cats_dogs_human_horses_cnn
I have done image classification of cats,dogs,humans,horses using CNN but still i am getting 67% accuracy on the validation data as i have done data augmentation also but still the accuray is too low. Can you please tell why this is happening?

#55

There need to some changes made :

  1. We use CNN’s to get information from images , hence the number of filters you use are directly proportional to the information you are gaining from the data. Hence , you need to increase number of filters in such a manner that you don’t over fit over the data. So , just change those to
    8 -> 16 -> 32 - > 64

  2. As your dataset is very small , hence you need to generate more images for it. Search about .flow() function of ImageDataGenerator . Use it to create more augmented images and store them in the same directory manner to use them further for training.

  3. Add Dropouts between dense layers.

I hope this will help You.
Thank You :slightly_smiling_face:.

#56


reduces to 54%.

#57

okay let me try it once.
Can you please share your dataset on the same github link.

#58

Bhaiya dataset is too large it cant be share. It contains total 808 images.

#59

I can whatsapp you the dataset?

#60

no dont whatsapp it.
Don’t worry , i will out some way.