function ret = cd1(rbm_w, visible_data)
% <rbm_w> is a matrix of size <number of hidden units> by <number of visible units>
% <visible_data> is a (possibly but not necessarily binary) matrix of size <number of visible units> by <number of data cases>
% The returned value is the gradient approximation produced by CD-1. It's of the same shape as <rbm_w>.


    % This is needed for questions up to question #9
    %v0 = visible_data;
    %

    v0 = sample_bernoulli(visible_data);
    
    h0p = visible_state_to_hidden_probabilities(rbm_w, v0);
    h0 = sample_bernoulli(h0p);

    v1p = hidden_state_to_visible_probabilities(rbm_w, h0);
    v1 = sample_bernoulli(v1p);

    h1p = visible_state_to_hidden_probabilities(rbm_w, v1);
    
    % h1 = sample_bernoulli(h1p); % We don't need this. But instead we could use
    % probabilities.

    ret = configuration_goodness_gradient(v0,h0) .- configuration_goodness_gradient(v1,h1p);

end