provide puthon coding

Question 3) Extreme Learning Machine: (20 points) We will now classify the Fisher Iris data using an extreme learning machine (ELM) with 4 input neurons, 40 hidden neurons, and 3 output neurons. This network will need two separate weight matrices; the first weight matrix, w1, projects the four input neurons (plus a bias) onto the 40 hidden neurons, while the second weight matrix, w2, projects the 40 hidden neurons (plus a bias) onto the three output neurons. In the ELM, the first weight matrix remains fixed and is never updated during training. Only the second weight matrix will be updated, using the online delta rule. In the code provided you will be responsible for implementing the forward pass xi=(wi01+j=14wij1uj)y^k(p)=(wk02+i=140wki2xi), where j indexes the input neurons, i indexes the hidden neurons, and k indexes the output neurons. The backward pass is computed by wki2=(y^k(p)yk(p))xi. where is the learning rate, y^(p) is the predicted output, and y(p) is the target output. For the bias term wk02, set x0=1. (a) Plot the train and test accuracy vs. the number of epochs. Report the final accuracy of the ELM network. (b) Repeat the same process using only two hidden neurons (instead of 40) and plot the accuracy vs. the number of epochs. (c) Graduate Students: Study how the networks accuracy changes with different weight initialization techniques: i) normal vs. uniform distribution, ii) only positive weights vs. bipolar weights, iii) adding sparsity to the weights (set 25%,50%,75%,90% of weights to 0 before training), and iv) varying the weight range (i.e. {1,1},{0.01,0.01}). Report final accuracies of your model in a table for each different scenario