nuget
api documents
Gibbs sampler for estimating the best assignments of topics for words and documents in a corpus. The algorithm is introduced in Tom Griffiths' paper "Gibbs sampling in the generative model of Latent Dirichlet Allocation" (2002). Gibbs sampler采样算法的实现
@author heinrich
00 Remarks
01 Syntax
02 Methods
| Name | Overloads | Summary |
|---|---|---|
| .ctor | 1 | Initialise the Gibbs sampler with data. |
| initialState | 1 | Initialisation: Must start with an assignment of observations to topics ? Many alternatives are possible, I chose to perform random assignments with equal probabilities 随机初始化状态 |
| gibbs | 2 | Main method: Select initial state ? Repeat a large number of times: 1. |
| sampling | 1 | for all z_i do chain data sampling |
| update_params | 1 | Add to the statistics the values of theta and phi for the current state. |
| configure | 1 | Configure the gibbs sampler 配置采样器 |
03 Properties
04 Fields
| Name | Overloads | Summary |
|---|---|---|
| documents | 1 | document data (term lists) 文档 |
| V | 1 | vocabulary size 词表大小 |
| alpha | 1 | Dirichlet parameter (document--topic associations) 文档——主题参数 |
| beta | 1 | Dirichlet parameter (topic--term associations) 主题——词语参数 |
| z | 1 | topic assignments for each word. |
| nw | 1 | cwt[i][j] number of instances of word i (term?) assigned to topic j. |
| nd | 1 | na[i][j] number of words in document i assigned to topic j. |
| nwsum | 1 | nwsum[j] total number of words assigned to topic j. |
| ndsum | 1 | nasum[i] total number of words in document i. |
| thetasum | 1 | cumulative statistics of theta theta的累积量 |
| phisum | 1 | cumulative statistics of phi phi的累积量 |
| numstats | 1 | size of statistics 样本容量 |
| THIN_INTERVAL | 1 | sampling lag (?) 多久更新一次统计量 |
| BURN_IN | 1 | burn-in period 收敛前的迭代次数 |
| ITERATIONS | 1 | max iterations 最大迭代次数 |
| SAMPLE_LAG | 1 | sample lag (if -1 only one sample taken) 最后的模型个数（取收敛后的n个迭代的参数做平均可以使得模型质量更高） |
05 Members
Int32()(), Int32, Action(Of Object))Initialise the Gibbs sampler with data. 用数据初始化采样器
| Name | Type | Description |
|---|---|---|
documents | Int32()() | 文档 |
V | Int32 | vocabulary size 词表大小 |
Int32)Initialisation: Must start with an assignment of observations to topics ? Many alternatives are possible, I chose to perform random assignments with equal probabilities 随机初始化状态
| Name | Type | Description |
|---|---|---|
K | Int32 | number of topics K个主题 |
Int32)with default cutoff alpha=2.0 and beta=0.5
| Name | Type | Description |
|---|---|---|
K | Int32 | - |
Int32, Double, Double)Main method: Select initial state ? Repeat a large number of times: 1. Select an element 2. Update conditional on other elements. If appropriate, output summary for each run. 采样
| Name | Type | Description |
|---|---|---|
K | Int32 | number of topics 主题数 |
alpha | Double | symmetric prior parameter on document--topic associations 对称文档——主题先验概率？ |
beta | Double | symmetric prior parameter on topic--term associations 对称主题——词语先验概率？ |
Int32)for all z_i do chain data sampling
| Name | Type | Description |
|---|---|---|
zi | Int32 | - |
Add to the statistics the values of theta and phi for the current state. 更新参数
Int32, Int32, Int32, Int32)Configure the gibbs sampler 配置采样器
| Name | Type | Description |
|---|---|---|
iterations | Int32 | number of total iterations |
burnIn | Int32 | number of burn-in iterations |
thinInterval | Int32 | update statistics interval |
sampleLag | Int32 | sample interval (-1 for just one sample at the end) |
number of topics 主题数目
Retrieve estimated document--topic associations. If sample lag > 0 then the mean value of all sampled statistics for theta[][] is taken. 获取文档——主题矩阵
theta multinomial mixture of document topics (M x K)
Retrieve estimated topic--word associations. If sample lag > 0 then the mean value of all sampled statistics for phi[][] is taken. 获取主题——词语矩阵
phi multinomial mixture of topic words (K x V)
document data (term lists) 文档
vocabulary size 词表大小
Dirichlet parameter (document--topic associations) 文档——主题参数
Dirichlet parameter (topic--term associations) 主题——词语参数
topic assignments for each word. 每个词语的主题 z[i][j] := 文档i的第j个词语的主题编号
cwt[i][j] number of instances of word i (term?) assigned to topic j. 计数器，nw[i][j] := 词语i归入主题j的次数
na[i][j] number of words in document i assigned to topic j. 计数器，nd[i][j] := 文档[i]中归入主题j的词语的个数
nwsum[j] total number of words assigned to topic j. 计数器，nwsum[j] := 归入主题j词语的个数
nasum[i] total number of words in document i. 计数器,ndsum[i] := 文档i中全部词语的数量
cumulative statistics of theta theta的累积量
cumulative statistics of phi phi的累积量
size of statistics 样本容量
sampling lag (?) 多久更新一次统计量
burn-in period 收敛前的迭代次数
max iterations 最大迭代次数
sample lag (if -1 only one sample taken) 最后的模型个数（取收敛后的n个迭代的参数做平均可以使得模型质量更高）