Just kidding, packrat is too confusing

.Rprofile

@@ -30,6 +30,3 @@ rm(RPROJ)
 # OR
 # R <- find_rstudio_root_file
 # R("Data", "data_raw")
-#### -- Packrat Autoloader (version 0.4.8-1) -- ####
-source("packrat/init.R")
-#### -- End Packrat Autoloader -- ####

.gitignore

@@ -28,10 +28,6 @@ Icon?
 .Rapp.history
 .Rproj.user
 
-# Packrat
-packrat/lib*/
-packrat/src/
-
 # --------
 # Python
 # --------

Analysis/ingo_survey/.Rprofile

@@ -1,5 +1,3 @@
 RPROJ <- list(PROJHOME = normalizePath("../.."))
 attach(RPROJ)
 rm(RPROJ)
-
-source("../../packrat/init.R", chdir = TRUE)

Analysis/ingo_survey/sandbox.R

@@ -1,85 +1,85 @@
-# # LDA
-# library(topicmodels)
-# thing_tm <- LDA(dtm.issues, k=5, method="Gibbs",
-#                 control = list(seed = 1234, burnin = 1000,
-#                                thin = 100, iter = 1000))
-# thing_tm1 <- LDA(dtm.issues, k=3, method="VEM",
-#                  control = list(seed=1234, estimate.alpha=FALSE))
-# topics(thing_tm1, 1)
-# terms(thing_tm1, 10)
-# 
-# topics.imputed <- data_frame(topic = 1:4,
-#                              title = c("Human rights, policy, and research",
-#                                        "Health and development", 
-#                                        "Health and advocacy",
-#                                        "Advocacy"))
-# 
-# issue.topics <- issue.corpus.df %>%
-#   mutate(topic = topics(thing_tm1, 1)) %>%
-#   left_join(topics.imputed, by="topic")
-# document.topics <- topics(thing_tm1, 1)
-# 
-# VEM_fixed = LDA(JSS_dtm, k = k,
-#                 + control = list(estimate.alpha = FALSE, seed = SEED))
-# 
-# # Latent semantic analysis
-# # https://meefen.github.io/blog/2013/03/11/analyze-text-similarity-in-r-latent-semantic-analysis-and-multidimentional-scaling/
-# library(lsa)
-# tdm.lsa <- lw_bintf(tdm.issues.mat) * gw_idf(tdm.issues.mat)
-# lsa.space <- lsa(tdm.lsa)
-# dist.mat.lsa <- dist(t(as.textmatrix(lsa.space)))  # compute distance matrix
-# 
-# fit <- cmdscale(dist.mat.lsa, eig = TRUE, k = 2)
-# points <- bind_cols(issue.corpus.df, 
-#                     data_frame(x = fit$points[, 1], y = fit$points[, 2]))
-# ggplot(points, aes(x = x, y = y)) + 
-#   geom_point()
-# 
-# # K-means clustering
-# # http://randyzwitch.com/rsitecatalyst-k-means-clustering/
-# # Find optimal number of k clusters
-# # http://stackoverflow.com/a/15376462/120898
-# # Different cost values
-# possible.clusters <- 1:25
-# 
-# # Sum of squared error for kmeans + elbow method (look for where slope changes)
-# kmeans.cost <- possible.clusters %>%
-#   map(function(i) kmeans(x=tdm.issues, centers=i, iter.max=100)$tot.withinss)
-# 
-# # Largest average silhouette width for pam (look for max avg.width)
-# pam.cost <- possible.clusters %>%
-#   map(function(i) pam(x=tdm.issues, k=i)$silinfo$avg.width)
-# 
-# # Combine costs for each algorithm into one dataframe
-# cost.df <- data_frame(cluster = possible.clusters, 
-#                       kmeans.cost = unlist(kmeans.cost),
-#                       pam.cost = c(NA, unlist(pam.cost))) %>%
-#   mutate(possible.group = case_when(
-#     .$cluster < 5 ~ 1,
-#     .$cluster >=5 & .$cluster <= 12 ~ 2,
-#     .$cluster > 12 ~ 3
-#   )) %>%
-#   mutate(possible.group = factor(possible.group))
-# 
-# # kmeans
-# ggplot(cost.df, aes(x=cluster, y=kmeans.cost)) + 
-#   geom_line() +
-#   geom_smooth(aes(colour=possible.group), method="lm", se=FALSE) +
-#   labs(x="Clusters", y="Within-cluster sum of squares",
-#        title="Optimal cluster size (kmeans)") + 
-#   guides(colour=FALSE) +
-#   theme_ath()
-# 
-# # pam
-# ggplot(na.omit(cost.df), aes(x=cluster, y=pam.cost)) +
-#   geom_line() +
-#   coord_cartesian(xlim=c(0, 25)) +
-#   labs(x="Clusters", y="Average silouette width",
-#        title="Optimal cluster size (pam)") +
-#   theme_ath()
-# 
-# # Create dataframe of clusters and term frequency
-# final.cluster <- pam(tdm.issues, k=4)$clustering
-# issues.clustered <- data_frame(cluster = final.cluster,
-#                                term = names(final.cluster)) %>%
-#   right_join(term.frequency, by="term")
+# LDA
+library(topicmodels)
+thing_tm <- LDA(dtm.issues, k=5, method="Gibbs",
+                control = list(seed = 1234, burnin = 1000,
+                               thin = 100, iter = 1000))
+thing_tm1 <- LDA(dtm.issues, k=3, method="VEM",
+                 control = list(seed=1234, estimate.alpha=FALSE))
+topics(thing_tm1, 1)
+terms(thing_tm1, 10)
+
+topics.imputed <- data_frame(topic = 1:4,
+                             title = c("Human rights, policy, and research",
+                                       "Health and development", 
+                                       "Health and advocacy",
+                                       "Advocacy"))
+
+issue.topics <- issue.corpus.df %>%
+  mutate(topic = topics(thing_tm1, 1)) %>%
+  left_join(topics.imputed, by="topic")
+document.topics <- topics(thing_tm1, 1)
+
+VEM_fixed = LDA(JSS_dtm, k = k,
+                + control = list(estimate.alpha = FALSE, seed = SEED))
+
+# Latent semantic analysis
+# https://meefen.github.io/blog/2013/03/11/analyze-text-similarity-in-r-latent-semantic-analysis-and-multidimentional-scaling/
+library(lsa)
+tdm.lsa <- lw_bintf(tdm.issues.mat) * gw_idf(tdm.issues.mat)
+lsa.space <- lsa(tdm.lsa)
+dist.mat.lsa <- dist(t(as.textmatrix(lsa.space)))  # compute distance matrix
+
+fit <- cmdscale(dist.mat.lsa, eig = TRUE, k = 2)
+points <- bind_cols(issue.corpus.df, 
+                    data_frame(x = fit$points[, 1], y = fit$points[, 2]))
+ggplot(points, aes(x = x, y = y)) + 
+  geom_point()
+
+# K-means clustering
+# http://randyzwitch.com/rsitecatalyst-k-means-clustering/
+# Find optimal number of k clusters
+# http://stackoverflow.com/a/15376462/120898
+# Different cost values
+possible.clusters <- 1:25
+
+# Sum of squared error for kmeans + elbow method (look for where slope changes)
+kmeans.cost <- possible.clusters %>%
+  map(function(i) kmeans(x=tdm.issues, centers=i, iter.max=100)$tot.withinss)
+
+# Largest average silhouette width for pam (look for max avg.width)
+pam.cost <- possible.clusters %>%
+  map(function(i) pam(x=tdm.issues, k=i)$silinfo$avg.width)
+
+# Combine costs for each algorithm into one dataframe
+cost.df <- data_frame(cluster = possible.clusters, 
+                      kmeans.cost = unlist(kmeans.cost),
+                      pam.cost = c(NA, unlist(pam.cost))) %>%
+  mutate(possible.group = case_when(
+    .$cluster < 5 ~ 1,
+    .$cluster >=5 & .$cluster <= 12 ~ 2,
+    .$cluster > 12 ~ 3
+  )) %>%
+  mutate(possible.group = factor(possible.group))
+
+# kmeans
+ggplot(cost.df, aes(x=cluster, y=kmeans.cost)) + 
+  geom_line() +
+  geom_smooth(aes(colour=possible.group), method="lm", se=FALSE) +
+  labs(x="Clusters", y="Within-cluster sum of squares",
+       title="Optimal cluster size (kmeans)") + 
+  guides(colour=FALSE) +
+  theme_ath()
+
+# pam
+ggplot(na.omit(cost.df), aes(x=cluster, y=pam.cost)) +
+  geom_line() +
+  coord_cartesian(xlim=c(0, 25)) +
+  labs(x="Clusters", y="Average silouette width",
+       title="Optimal cluster size (pam)") +
+  theme_ath()
+
+# Create dataframe of clusters and term frequency
+final.cluster <- pam(tdm.issues, k=4)$clustering
+issues.clustered <- data_frame(cluster = final.cluster,
+                               term = names(final.cluster)) %>%
+  right_join(term.frequency, by="term")

Analysis/ngo_regs_regime_stability/.Rprofile

@@ -1,5 +1,3 @@
 RPROJ <- list(PROJHOME = normalizePath("../.."))
 attach(RPROJ)
 rm(RPROJ)
-
-source("../../packrat/init.R", chdir = TRUE)