Day 7-BackPropogration In Recurrent Neural Network And NLP Application|Krish Naik

[@simranvolunesia]

1st distribution: Pareto Distribution (power law probability distribution, talks about 80-20 rule i.e. 80% of outcomes are result of 20% causes) Convert pareto distribution to normal distribution: using box cox transformation (it converts to nearly normal distribution, formula is: x_new = log(x_old) if lambda = 0 x_new = (x_old**lambda - 1)/lambda otherwise) Check if distribution is normal gaussian distribution or not: Draw QQ plots (Quantile-Quantile plot) where we draw actual quantiles for data and theoretical quantiles and then look at deviation. What is standard normal distribution? Normal distribution with mean 0 and std 1 Plot A: Right Skewed mean > median > mode Plot B: Left Skewed mode > median > mean Difference between fit_transform and transform fit_transform do two operations on input data in one go which are fit the transformer as per the calculations on input data and then apply those calculations to input data transform: will only apply calculations on input data as per transformer. When we use fit_predict and what is it? fit_predict fits the model as per input data and later on make a prediction as well using trained model. We generally use it in clustering algorithms like DBSCAN where we can only do fit_predict or fit. Predict is not possible there. What is difference between standardization and normalization? 1. Normalization is basically min max scaling 2. Values belong to [0,1] or [-1,1] 3. Affected by outliers 4. Useful when we don't know about distribution 1. Standardization converts data in a manner that mean of transformed data is 0 and std=1 2. No bound on values 3. Not much affected by outliers 4. Useful when data distribution is normal gaussian distribution. PS: Do correct me if I went wrong somewhere.

[@vipulpatil7687]

Thanks a lot

[@devkumaracharyaiitbombay5341]

nice work

[@pankajkumarbarman765]

Thank you sir for this Amazing lecture 🥰🥰🥰🥰

[@mainuddinali9561]

great

[@mrityunjayupadhyay7332]

Amazing explanation

[@sandipansarkar9211]

finished watching

[@nik7867]

RNN’s can work on sentences having small length as there hasn’t been any significant change in weights with respect to distance among the words and the context is preserved . Am I correct ?

[@aliali-sm3dq]

Thank you sir. would you please put all of interview question in seperated video?

[@pranit7266]

Sir please upload the class material

[@pranjalsingh1287]

do make video of lstm and rnn with the context of sound files processing

[@pranavrane5679]

Not able to find course material on ineuron. Could you please help?

[@himanshuagarwal3788]

sir, can you please tell what product you use for writing & presenting that in screen...

[@TheNishi42]

Back propagation always done using gradient descent only ?

[@solo-ue4ii]

NOPE IG

[@MoosaMemon.]

No, there are other better optimizers out there. The most popular and widely used optimizer for almost every scenario is "Adam's Optimizer".

[@HarshaVardhan-hd2ts]

Since in RNN , the weights are same when intialized , is that the same case for CNN also ? Can someone help me with this

[@mengli6949]

what if the hidden output o' of different dimension than the x input? if x input is 300 dim, how do we choose the length of o'?

[@dhanushnayak1674]

hi krish r u fyn..looking dull.