Merge pull request #137 from openvenues/fix_address_parser_train

albarrentine · web-flow · commit bcf6b3cc683d · 2016-12-12T11:54:16.000-05:00
Fix address_parser_train
diff --git a/src/address_parser_io.c b/src/address_parser_io.c
@@ -31,7 +31,7 @@ bool address_parser_data_set_tokenize_line(address_parser_data_set_t *data_set,
     uint32_t i = 0;
     char *str = NULL;
 
-    cstring_array *pairs = cstring_array_split(input, " ", 1, &count);
+    cstring_array *pairs = cstring_array_split_ignore_consecutive(input, " ", 1, &count);
     size_t num_pairs = cstring_array_num_strings(pairs);
 
     char *label = NULL;
@@ -62,23 +62,57 @@ bool address_parser_data_set_tokenize_line(address_parser_data_set_t *data_set,
         }
 
         token.offset = pairs->indices->a[i];
-        token.len = last_separator_index;
+        size_t expected_len = last_separator_index;
 
-        scanner_t scanner = scanner_from_string(input + token.offset, token.len);
+        scanner_t scanner = scanner_from_string(input + token.offset, expected_len);
         token.type = scan_token(&scanner);
-        if (ADDRESS_PARSER_IS_SEPARATOR(token.type)) {
-            uint32_array_push(separators, ADDRESS_SEPARATOR_FIELD_INTERNAL);
-            continue;
-        } else if (ADDRESS_PARSER_IS_IGNORABLE(token.type)) {
-            // shouldn't happen but just in case
-            continue;
+        token.len = scanner.cursor - scanner.start;
+
+        if (token.len == expected_len) {
+            if (ADDRESS_PARSER_IS_SEPARATOR(token.type)) {
+                uint32_array_push(separators, ADDRESS_SEPARATOR_FIELD_INTERNAL);
+                continue;
+            } else if (ADDRESS_PARSER_IS_IGNORABLE(token.type)) {
+                // shouldn't happen but just in case
+                continue;
+            } else {
+                uint32_array_push(separators, ADDRESS_SEPARATOR_NONE);
+            }
+
+            cstring_array_add_string(labels, label);
+
+            token_array_push(tokens, token);
         } else {
-            uint32_array_push(separators, ADDRESS_SEPARATOR_NONE);
-        }
+            /* If normalizing the string turned one token into several e.g. ½ => 1/2
+               add all the tokens where offset = (token.offset + sub_token.offset)
+               with the same label as the parent.
+            */
+            token_array *sub_tokens = token_array_new();
+            if (sub_tokens == NULL) {
+                log_error("Error allocating sub-token array\n");
+                return false;
+            }
+            tokenize_add_tokens(sub_tokens, input + token.offset, expected_len, false);
+            for (size_t j = 0; j < sub_tokens->n; j++) {
+                token_t sub_token = sub_tokens->a[j];
+                // Add the offset of the parent "token"
+                sub_token.offset = token.offset + sub_token.offset;
+
+                if (ADDRESS_PARSER_IS_SEPARATOR(sub_token.type)) {
+                    uint32_array_push(separators, ADDRESS_SEPARATOR_FIELD_INTERNAL);
+                    continue;
+                } else if (ADDRESS_PARSER_IS_IGNORABLE(sub_token.type)) {
+                    continue;
+                } else {
+                    uint32_array_push(separators, ADDRESS_SEPARATOR_NONE);
+                }
+
+                cstring_array_add_string(labels, label);
+                token_array_push(tokens, sub_token);
+            }
 
-        cstring_array_add_string(labels, label);
+        }
 
-        token_array_push(tokens, token);
     })
 
     cstring_array_destroy(pairs);
diff --git a/src/address_parser_test.c b/src/address_parser_test.c
@@ -142,6 +142,14 @@ int main(int argc, char **argv) {
 
     log_info("address dictionary module loaded\n");
 
+    // Needs to load for normalization
+    if (!transliteration_module_setup(NULL)) {
+        log_error("Could not load transliteration module\n");
+        exit(EXIT_FAILURE);
+    }
+
+    log_info("transliteration module loaded\n");
+
     if (!geodb_module_setup(NULL)) {
         log_error("Could not load geodb dictionaries\n");
         exit(EXIT_FAILURE);
diff --git a/src/address_parser_train.c b/src/address_parser_train.c
@@ -7,6 +7,7 @@
 #include "file_utils.h"
 #include "geodb.h"
 #include "shuffle.h"
+#include "transliterate.h"
 
 #include "log/log.h"
 
@@ -450,6 +451,14 @@ int main(int argc, char **argv) {
 
     log_info("address dictionary module loaded\n");
 
+    // Needs to load for normalization
+    if (!transliteration_module_setup(NULL)) {
+        log_error("Could not load transliteration module\n");
+        exit(EXIT_FAILURE);
+    }
+
+    log_info("transliteration module loaded\n");
+
     if (!geodb_module_setup(NULL)) {
         log_error("Could not load geodb dictionaries\n");
         exit(EXIT_FAILURE);
diff --git a/src/string_utils.c b/src/string_utils.c
@@ -829,17 +829,23 @@ inline int64_t cstring_array_token_length(cstring_array *self, uint32_t i) {
     }
 }
 
-cstring_array *cstring_array_split(char *str, const char *separator, size_t separator_len, size_t *count) {
+static cstring_array *cstring_array_split_options(char *str, const char *separator, size_t separator_len, bool ignore_consecutive, size_t *count) {
     *count = 0;
     char_array *array = char_array_new_size(strlen(str));
 
+    bool last_was_separator = false;
+
     while (*str) {
         if ((separator_len == 1 && *str == separator[0]) || (memcmp(str, separator, separator_len) == 0)) {
-            char_array_push(array, '\0');
+            if (!ignore_consecutive || !last_was_separator) {
+                char_array_push(array, '\0');
+            }
             str += separator_len;
+            last_was_separator = true;
         } else {
             char_array_push(array, *str);
             str++;
+            last_was_separator = false;
         }
     }
     char_array_push(array, '\0');
@@ -850,6 +856,17 @@ cstring_array *cstring_array_split(char *str, const char *separator, size_t sepa
     return string_array;
 }
 
+
+cstring_array *cstring_array_split(char *str, const char *separator, size_t separator_len, size_t *count) {
+    return cstring_array_split_options(str, separator, separator_len, false, count);
+}
+
+
+cstring_array *cstring_array_split_ignore_consecutive(char *str, const char *separator, size_t separator_len, size_t *count) {
+    return cstring_array_split_options(str, separator, separator_len, true, count);
+}
+
+
 cstring_array *cstring_array_split_no_copy(char *str, char separator, size_t *count) {
     *count = 0;
     char *ptr = str;
diff --git a/src/string_utils.h b/src/string_utils.h
@@ -180,6 +180,8 @@ char **cstring_array_to_strings(cstring_array *self);
 
 // Split on delimiter
 cstring_array *cstring_array_split(char *str, const char *separator, size_t separator_len, size_t *count);
+// Split on delimiter, ignore multiple consecutive delimiters
+cstring_array *cstring_array_split_ignore_consecutive(char *str, const char *separator, size_t separator_len, size_t *count);
 
 // Split on delimiter by replacing (single character) separator with the NUL byte in the original string
 cstring_array *cstring_array_split_no_copy(char *str, char separator, size_t *count);
diff --git a/src/tokens.c b/src/tokens.c
@@ -6,7 +6,6 @@ tokenized_string_t *tokenized_string_new(void) {
     self->str = NULL;
     self->strings = cstring_array_new();
     self->tokens = token_array_new();
-
     return self;
 }
 
@@ -21,7 +20,11 @@ tokenized_string_t *tokenized_string_new_size(size_t len, size_t num_tokens) {
 
 inline tokenized_string_t *tokenized_string_new_from_str_size(char *src, size_t len, size_t num_tokens) {
     tokenized_string_t *self = tokenized_string_new_size(len, num_tokens);
-    self->str = src;
+    self->str = strndup(src, len);
+    if (self->str == NULL) {
+        tokenized_string_destroy(self);
+        return NULL;
+    }
     return self;
 }
 
@@ -38,7 +41,11 @@ void tokenized_string_add_token(tokenized_string_t *self, const char *src, size_
 
 tokenized_string_t *tokenized_string_from_tokens(char *src, token_array *tokens, bool copy_tokens) {
     tokenized_string_t *self = malloc(sizeof(tokenized_string_t));
-    self->str = src;
+    self->str = strdup(src);
+    if (self->str == NULL) {
+        tokenized_string_destroy(self);
+        return NULL;
+    }
     self->strings = cstring_array_new_size(strlen(src) + tokens->n);
     if (copy_tokens) {
         self->tokens = token_array_new_copy(tokens, tokens->n);
@@ -48,7 +55,7 @@ tokenized_string_t *tokenized_string_from_tokens(char *src, token_array *tokens,
 
     token_t token;
 
-    for (int i = 0; i < tokens->n; i++) {
+    for (size_t i = 0; i < tokens->n; i++) {
         token = tokens->a[i];
         cstring_array_add_string_len(self->strings, src + token.offset, token.len);
     }
