I am trying to sample from the posterior resulting of this (Logistic Regression) model specification:
lower = 0.01 upper = 1 with pm.Model() as customer_model: beta_0 = pm.Uniform('beta_0', lower=lower, upper= upper) beta_feature_one = pm.Uniform( 'beta_feature_one', lower = X_train['feature_one'].min(), upper = X_train['feature_one'].max() ) beta_feature_two= pm.Uniform( 'beta_feature_two', lower = X_train['feature_two'].min(), upper = X_train['feature_two'].max() ) beta_feature_three = pm.Uniform( 'beta_feature_three', lower = X_train['feature_three'].min(), upper = X_train['feature_three'].max(), ) beta_feature_four = pm.TruncatedNormal( 'beta_feature_four', mu = 10, sigma = 44, lower = 0 ) beta_feature_five = pm.Normal( 'beta_feature_five', mu = 598, sigma = 340 ) beta_feature_six = pm.Normal( 'beta_feature_six', mu = 32180, sigma = 15687 ) lin_comb = ( beta_0 + beta_feature_one * X_train['feature_one'] + beta_feature_two* X_train['feature_two'] + beta_feature_three * X_train['feature_three'] + beta_feature_four * X_train['feature_four'] + beta_feature_five * X_train['feature_five'] + beta_feature_six * X_train['feature_six'] ) p = pm.Deterministic('p', pm.math.sigmoid(lin_comb)) likelihood = pm.Bernoulli( 'observed_default', p = p, observed = y_train ) trace = pm.sample()
I am unsure about which values to pass to
pm.sample() so I decided to keep the default values (after passing other values first). Also, I cannot discern any distribution after plotting selected features, so I have decided to assume a uniform distribution for them. Finally, I am aware of the importance of scaling features first, but decided to run this version of the project without scaling.
This is the error message I get
Cannot execute code, session has been disposed. Please try restarting the Kernel. The Kernel crashed while executing code in the the current cell or a previous cell.(...)
I have restarted the Kernel and cleared the notebook prior to running it again (but getting the same error message). I am running the model in
What am I missing in this model specification? Could it be that it is not the model itself, but that I need to scale the features first?
Thanks in advance!