Python:key error

user_1372560 0
Viewed 356

I'm using scikit-learn for GMM training and am trying to vary the number of mixture components. I ran into problems, which were fixed here.

I ended up with this code:

from sklearn.mixture import GMM

class_names = ['name1','name2','name3']
covs =  ['spherical', 'diagonal', 'tied', 'full']
num_comp = [1,2,3]
models = dict()

for c in class_names:
    for cov in covs:
        models[c,cov] = values = []
        for num in num_comp:
            values.append(GMM(n_components=num,covariance_type=cov, init_params='wmc',n_init=1, n_iter=10))
print models

I'm using the model dict() later on again and am running into problems because I don't know how to work with the dict.

training_data={'name1':'path2data1', 'name2': path2data2, 'name3': path2data3}

for cov in covs:
    for c in class_names:
        for num in num_comp:
            models[c][cov].fit(training_data[c])

I get the error "KeyError:'name1'"

I tried models[c,cov].fit(training_data[c]) as well but then I get the error "AttributeError: 'list' object has no attribute 'fit'"

Thanks in advance for any tips!

scikit-learn dictionary python
1 Answers

Here you're assign value to dict by tuple (c, cov is equivalent to (c, cov)):

models[c,cov] = values = []

So, here you should access to dict values by tuple to:

for cov in covs:
    for c in class_names:
        for num in num_comp:
            models[(c, cov)].fit(training_data[c])

AttributeError throwing because you actually access to list: models[(c,cov)] is list that you created before (values)

update for comment, try this:

for c in class_names:
    models[c] = {}
    for cov in covs:
        models[c][cov] = values = []
        for num in num_comp:
            values.append(GMM(n_components=num,covariance_type=cov, init_params='wmc',n_init=1, n_iter=10))

but here you will get list in models[c][cov] anyway and should iterate it:

for cov in covs:
    for c in class_names:
        for num in num_comp:
            for f in models[c][cov]
                f.fit(training_data[c])
user_10998760
Related