问题
I have a dataframe as follows. I need to do ANOVA on this between three conditions. The dataframe looks like:
data0 = pd.DataFrame({'Names': ['CTA15', 'CTA15', 'AC007', 'AC007', 'AC007','AC007'],
'value': [22, 22, 2, 2, 2,5],
'condition':['NON', 'NON', 'YES', 'YES', 'RE','RE']})
I need to do ANOVA test between YES and NON, NON and RE and YES and RE, conditions from conditions for Names. I know I could do it like this,
NON=df.query('condition =="NON"and Names=="CTA15"')
no=df.value
YES=df.query('condition =="YES"and Names=="CTA15"')
Y=YES.value
Then perform one way ANOVA as following,
from scipy import stats
f_val, p_val = stats.f_oneway(no, Y)
print ("One-way ANOVA P =", p_val )
But would be great if there is any elegant solution as my initial data frame is big and has many names and conditions to compare between
回答1:
Consider the following sample DataFrame:
df = pd.DataFrame({'Names': np.random.randint(1, 10, 1000),
'value': np.random.randn(1000),
'condition': np.random.choice(['NON', 'YES', 'RE'], 1000)})
df.head()
Out:
Names condition value
0 4 RE 0.844120
1 4 NON -0.440285
2 5 YES 0.559497
3 4 RE 0.472425
4 9 YES 0.205906
The following groups the DataFrame by Names, and then passes each condition group to ANOVA:
import scipy.stats as ss
for name_group in df.groupby('Names'):
samples = [condition[1] for condition in name_group[1].groupby('condition')['value']]
f_val, p_val = ss.f_oneway(*samples)
print('Name: {}, F value: {:.3f}, p value: {:.3f}'.format(name_group[0], f_val, p_val))
Name: 1, F value: 0.138, p value: 0.871
Name: 2, F value: 1.458, p value: 0.237
Name: 3, F value: 0.742, p value: 0.479
Name: 4, F value: 2.718, p value: 0.071
Name: 5, F value: 0.255, p value: 0.776
Name: 6, F value: 1.731, p value: 0.182
Name: 7, F value: 0.269, p value: 0.764
Name: 8, F value: 0.474, p value: 0.624
Name: 9, F value: 1.226, p value: 0.297
For post-hoc tests, you can use statsmodels (as explained here):
from statsmodels.stats.multicomp import pairwise_tukeyhsd
for name, grouped_df in df.groupby('Names'):
print('Name {}'.format(name), pairwise_tukeyhsd(grouped_df['value'], grouped_df['condition']))
Name 1 Multiple Comparison of Means - Tukey HSD,FWER=0.05 ============================================ group1 group2 meandiff lower upper reject -------------------------------------------- NON RE 0.0086 -0.5129 0.5301 False NON YES 0.0084 -0.4817 0.4986 False RE YES -0.0002 -0.5217 0.5214 False -------------------------------------------- Name 2 Multiple Comparison of Means - Tukey HSD,FWER=0.05 ============================================ group1 group2 meandiff lower upper reject -------------------------------------------- NON RE -0.0089 -0.5299 0.5121 False NON YES 0.083 -0.4182 0.5842 False RE YES 0.0919 -0.4008 0.5846 False -------------------------------------------- Name 3 Multiple Comparison of Means - Tukey HSD,FWER=0.05 ============================================ group1 group2 meandiff lower upper reject -------------------------------------------- NON RE 0.2401 -0.3136 0.7938 False NON YES 0.2765 -0.2903 0.8432 False RE YES 0.0364 -0.5052 0.578 False -------------------------------------------- Name 4 Multiple Comparison of Means - Tukey HSD,FWER=0.05 ============================================ group1 group2 meandiff lower upper reject -------------------------------------------- NON RE 0.0894 -0.5825 0.7613 False NON YES -0.0437 -0.7418 0.6544 False RE YES -0.1331 -0.6949 0.4287 False -------------------------------------------- Name 5 Multiple Comparison of Means - Tukey HSD,FWER=0.05 ============================================ group1 group2 meandiff lower upper reject -------------------------------------------- NON RE -0.4264 -0.9495 0.0967 False NON YES 0.0439 -0.4264 0.5142 False RE YES 0.4703 -0.0155 0.9561 False -------------------------------------------- Name 6 Multiple Comparison of Means - Tukey HSD,FWER=0.05 ============================================ group1 group2 meandiff lower upper reject -------------------------------------------- NON RE 0.0649 -0.4971 0.627 False NON YES -0.406 -0.9405 0.1285 False RE YES -0.4709 -1.0136 0.0717 False -------------------------------------------- Name 7 Multiple Comparison of Means - Tukey HSD,FWER=0.05 ============================================ group1 group2 meandiff lower upper reject -------------------------------------------- NON RE 0.3111 -0.2766 0.8988 False NON YES -0.1664 -0.7314 0.3987 False RE YES -0.4774 -1.0688 0.114 False -------------------------------------------- Name 8 Multiple Comparison of Means - Tukey HSD,FWER=0.05 ============================================ group1 group2 meandiff lower upper reject -------------------------------------------- NON RE -0.0224 -0.668 0.6233 False NON YES 0.0119 -0.668 0.6918 False RE YES 0.0343 -0.6057 0.6742 False -------------------------------------------- Name 9 Multiple Comparison of Means - Tukey HSD,FWER=0.05 ============================================ group1 group2 meandiff lower upper reject -------------------------------------------- NON RE -0.2414 -0.7792 0.2963 False NON YES 0.0696 -0.5746 0.7138 False RE YES 0.311 -0.3129 0.935 False
来源：https://stackoverflow.com/questions/44065573/anova-for-groups-within-a-dataframe-using-scipy