skimtoken 0.2.1__tar.gz → 0.2.2__tar.gz

{skimtoken-0.2.1 → skimtoken-0.2.2}/Cargo.lock

@@ -392,7 +392,7 @@ dependencies = [
 
 [[package]]
 name = "skimtoken"
-version = "0.2.1"
+version = "0.2.2"
 dependencies = [
  "atty",
  "clap",

{skimtoken-0.2.1 → skimtoken-0.2.2}/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "skimtoken"
-version = "0.2.1"
+version = "0.2.2"
 edition = "2021"
 authors = ["masaishi <mwishiha@ucsc.edu>"]
 license = "MIT"

{skimtoken-0.2.1 → skimtoken-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: skimtoken
-Version: 0.2.1
+Version: 0.2.2
 License-File: LICENSE
 Summary: Fast token count estimation library
 Home-Page: https://github.com/masaishi/skimtoken
@@ -21,23 +21,16 @@ A lightweight, fast token count estimation library written in Rust with Python b
 [![Crates.io](https://img.shields.io/crates/v/skimtoken)](https://crates.io/crates/skimtoken)
 [![License](https://img.shields.io/github/license/masaishi/skimtoken)](https://github.com/masaishi/skimtoken/blob/main/LICENSE)
 
-## ⚠️ Current Limitations
-
-**This library is currently in early beta and has significant accuracy issues:**
-
-- **Multilingual method**: Takes 1.13x longer than tiktoken due to inefficient implementation
-- **Overall accuracy**: 15.11% error rate, which is too high for most use cases
-
 
 ## Why skimtoken?
 
-**The Problem**: [tiktoken](https://github.com/openai/tiktoken) is great for precise tokenization, but requires ~60MB of memory just to count tokens - problematic for memory-constrained environments.
+**The Problem**: [tiktoken](https://github.com/openai/tiktoken) is great for precise tokenization, but requires ~59.6MB of memory just to count tokens - problematic for memory-constrained environments.
 
 **The Solution**: skimtoken estimates token counts using statistical patterns instead of loading entire vocabularies, achieving:
 
-- ✅ **64x less memory** (0.92MB vs 60MB)
-- ✅ **128x faster startup** (4ms vs 485ms) 
-- ❌ **1.13x slower execution** (5.51s vs 4.59s) for multilingual method
+- ✅ **65x less memory** (0.92MB vs 59.6MB)
+- ✅ **421x faster startup** (2.389ms vs 1,005ms)
+- ❌ **1.03x slowwer execute time** (6.689s vs 6.912s) for Multilingual single method
 - ❌ Trade-off: ~15.11% error rate vs exact counts
 
 ## Installation
@@ -100,32 +93,7 @@ print(f"Estimated tokens (multilingual): {token_count}")
 
 ### Large-Scale Benchmark (100k samples)
 
-Simple method (Just char length x coefficient):
-```
-Results:
-Total Samples: 100,726
-Total Characters: 13,062,391
-Mean RMSE: 38.4863 tokens
-Mean Error Rate: 21.63%
-
-┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━┓
-┃ Metric       ┃   tiktoken ┃  skimtoken ┃  Ratio ┃
-┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━┩
-│ Init Time    │ 0.481672 s │ 0.182308 s │ 0.378x │
-├──────────────┼────────────┼────────────┼────────┤
-│ Init Memory  │ 42.2386 MB │  0.0291 MB │ 0.001x │
-├──────────────┼────────────┼────────────┼────────┤
-│ Exec Time    │ 4.710224 s │ 0.805272 s │ 0.171x │
-├──────────────┼────────────┼────────────┼────────┤
-│ Exec Memory  │ 17.3251 MB │  0.8849 MB │ 0.051x │
-├──────────────┼────────────┼────────────┼────────┤
-│ Total Time   │ 5.191896 s │ 0.928758 s │ 0.190x │
-├──────────────┼────────────┼────────────┼────────┤
-│ Total Memory │ 59.5637 MB │  0.9214 MB │ 0.015x │
-└──────────────┴────────────┴────────────┴────────┘
-```
-
-Multilingual simple method:
+Multilingual single method:
 ```
 Results:
 Total Samples: 100,726
@@ -136,27 +104,35 @@ Mean Error Rate: 15.11%
 ┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━┓
 ┃ Metric       ┃   tiktoken ┃  skimtoken ┃  Ratio ┃
 ┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━┩
-│ Init Time    │ 0.815441 s │ 0.138714 s │ 0.170x │
+│ Init Time    │ 1.005490 s │ 0.002389 s │ 0.002x │
 ├──────────────┼────────────┼────────────┼────────┤
-│ Init Memory  │ 42.4791 MB │  0.1613 MB │ 0.004x │
+│ Init Memory  │ 42.2310 MB │  0.0265 MB │ 0.001x │
 ├──────────────┼────────────┼────────────┼────────┤
-│ Exec Time    │ 4.041857 s │ 5.380782 s │ 1.331x │
+│ Exec Time    │ 6.689203 s │ 6.911931 s │ 1.033x │
 ├──────────────┼────────────┼────────────┼────────┤
-│ Exec Memory  │ 17.3227 MB │  0.8950 MB │ 0.052x │
+│ Exec Memory  │ 17.3251 MB │  0.8950 MB │ 0.052x │
 ├──────────────┼────────────┼────────────┼────────┤
-│ Total Time   │ 4.857297 s │ 5.519496 s │ 1.136x │
+│ Total Time   │ 7.694694 s │ 6.914320 s │ 0.899x │
 ├──────────────┼────────────┼────────────┼────────┤
-│ Total Memory │ 59.8018 MB │  1.0563 MB │ 0.018x │
+│ Total Memory │ 59.5561 MB │  0.9215 MB │ 0.015x │
 └──────────────┴────────────┴────────────┴────────┘
 ```
 
+### Automated Benchmarks
+
+For up-to-date performance comparisons and detailed accuracy metrics across all methods, visit the [skimtoken_benchmark](https://github.com/masaishi/skimtoken_benchmark) repository. This automated benchmark suite:
+
+- Uses the CC-100 multilingual dataset (100k+ samples)
+- Provides language-specific accuracy breakdowns
+
 ## Available Methods
 
 | Method | Import | Memory | Error | Best For |
 |--------|---------|--------|-------|----------|
-| **Simple** | `from skimtoken.simple import estimate_tokens` | 0.8MB | ~21% | English text, minimum memory |
-| **Basic** | `from skimtoken.basic import estimate_tokens` | 0.8MB | ~27% | General use |
-| **Multilingual** | `from skimtoken.multilingual import estimate_tokens` | 0.9MB | ~15% | Non-English, mixed languages |
+| **Simple** | `from skimtoken.simple import estimate_tokens` | 1.0MB | ~21.63% | English text, minimum memory |
+| **Basic** | `from skimtoken.basic import estimate_tokens` | 0.9MB | ~27.05% | General use |
+| **Multilingual** | `from skimtoken.multilingual import estimate_tokens` | 0.9MB | ~15.93% | Non-English, mixed languages |
+| **Multilingual Simple** | `from skimtoken.multilingual_simple import estimate_tokens` | 0.9MB | ~15.11% | Fast multilingual estimation |
 
 ```python
 # Example: Choose method based on your needs
@@ -272,7 +248,7 @@ A: Beta = breaking changes possible.
 We are actively working to improve skimtoken's accuracy and performance:
 
 1. **Better estimation algorithms**: Moving beyond simple character multiplication to more sophisticated statistical models
-2. **Performance optimization**: Fixing the 60x slowdown in multilingual method
+2. **Performance optimization**: Further improving execution speed
 3. **Improved language support**: Better handling of non-English languages
 4. **Higher accuracy**: Targeting <10% error rate while maintaining low memory footprint
 

{skimtoken-0.2.1 → skimtoken-0.2.2}/README.md

@@ -8,23 +8,16 @@ A lightweight, fast token count estimation library written in Rust with Python b
 [![Crates.io](https://img.shields.io/crates/v/skimtoken)](https://crates.io/crates/skimtoken)
 [![License](https://img.shields.io/github/license/masaishi/skimtoken)](https://github.com/masaishi/skimtoken/blob/main/LICENSE)
 
-## ⚠️ Current Limitations
-
-**This library is currently in early beta and has significant accuracy issues:**
-
-- **Multilingual method**: Takes 1.13x longer than tiktoken due to inefficient implementation
-- **Overall accuracy**: 15.11% error rate, which is too high for most use cases
-
 
 ## Why skimtoken?
 
-**The Problem**: [tiktoken](https://github.com/openai/tiktoken) is great for precise tokenization, but requires ~60MB of memory just to count tokens - problematic for memory-constrained environments.
+**The Problem**: [tiktoken](https://github.com/openai/tiktoken) is great for precise tokenization, but requires ~59.6MB of memory just to count tokens - problematic for memory-constrained environments.
 
 **The Solution**: skimtoken estimates token counts using statistical patterns instead of loading entire vocabularies, achieving:
 
-- ✅ **64x less memory** (0.92MB vs 60MB)
-- ✅ **128x faster startup** (4ms vs 485ms) 
-- ❌ **1.13x slower execution** (5.51s vs 4.59s) for multilingual method
+- ✅ **65x less memory** (0.92MB vs 59.6MB)
+- ✅ **421x faster startup** (2.389ms vs 1,005ms)
+- ❌ **1.03x slowwer execute time** (6.689s vs 6.912s) for Multilingual single method
 - ❌ Trade-off: ~15.11% error rate vs exact counts
 
 ## Installation
@@ -87,32 +80,7 @@ print(f"Estimated tokens (multilingual): {token_count}")
 
 ### Large-Scale Benchmark (100k samples)
 
-Simple method (Just char length x coefficient):
-```
-Results:
-Total Samples: 100,726
-Total Characters: 13,062,391
-Mean RMSE: 38.4863 tokens
-Mean Error Rate: 21.63%
-
-┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━┓
-┃ Metric       ┃   tiktoken ┃  skimtoken ┃  Ratio ┃
-┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━┩
-│ Init Time    │ 0.481672 s │ 0.182308 s │ 0.378x │
-├──────────────┼────────────┼────────────┼────────┤
-│ Init Memory  │ 42.2386 MB │  0.0291 MB │ 0.001x │
-├──────────────┼────────────┼────────────┼────────┤
-│ Exec Time    │ 4.710224 s │ 0.805272 s │ 0.171x │
-├──────────────┼────────────┼────────────┼────────┤
-│ Exec Memory  │ 17.3251 MB │  0.8849 MB │ 0.051x │
-├──────────────┼────────────┼────────────┼────────┤
-│ Total Time   │ 5.191896 s │ 0.928758 s │ 0.190x │
-├──────────────┼────────────┼────────────┼────────┤
-│ Total Memory │ 59.5637 MB │  0.9214 MB │ 0.015x │
-└──────────────┴────────────┴────────────┴────────┘
-```
-
-Multilingual simple method:
+Multilingual single method:
 ```
 Results:
 Total Samples: 100,726
@@ -123,27 +91,35 @@ Mean Error Rate: 15.11%
 ┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━┓
 ┃ Metric       ┃   tiktoken ┃  skimtoken ┃  Ratio ┃
 ┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━┩
-│ Init Time    │ 0.815441 s │ 0.138714 s │ 0.170x │
+│ Init Time    │ 1.005490 s │ 0.002389 s │ 0.002x │
 ├──────────────┼────────────┼────────────┼────────┤
-│ Init Memory  │ 42.4791 MB │  0.1613 MB │ 0.004x │
+│ Init Memory  │ 42.2310 MB │  0.0265 MB │ 0.001x │
 ├──────────────┼────────────┼────────────┼────────┤
-│ Exec Time    │ 4.041857 s │ 5.380782 s │ 1.331x │
+│ Exec Time    │ 6.689203 s │ 6.911931 s │ 1.033x │
 ├──────────────┼────────────┼────────────┼────────┤
-│ Exec Memory  │ 17.3227 MB │  0.8950 MB │ 0.052x │
+│ Exec Memory  │ 17.3251 MB │  0.8950 MB │ 0.052x │
 ├──────────────┼────────────┼────────────┼────────┤
-│ Total Time   │ 4.857297 s │ 5.519496 s │ 1.136x │
+│ Total Time   │ 7.694694 s │ 6.914320 s │ 0.899x │
 ├──────────────┼────────────┼────────────┼────────┤
-│ Total Memory │ 59.8018 MB │  1.0563 MB │ 0.018x │
+│ Total Memory │ 59.5561 MB │  0.9215 MB │ 0.015x │
 └──────────────┴────────────┴────────────┴────────┘
 ```
 
+### Automated Benchmarks
+
+For up-to-date performance comparisons and detailed accuracy metrics across all methods, visit the [skimtoken_benchmark](https://github.com/masaishi/skimtoken_benchmark) repository. This automated benchmark suite:
+
+- Uses the CC-100 multilingual dataset (100k+ samples)
+- Provides language-specific accuracy breakdowns
+
 ## Available Methods
 
 | Method | Import | Memory | Error | Best For |
 |--------|---------|--------|-------|----------|
-| **Simple** | `from skimtoken.simple import estimate_tokens` | 0.8MB | ~21% | English text, minimum memory |
-| **Basic** | `from skimtoken.basic import estimate_tokens` | 0.8MB | ~27% | General use |
-| **Multilingual** | `from skimtoken.multilingual import estimate_tokens` | 0.9MB | ~15% | Non-English, mixed languages |
+| **Simple** | `from skimtoken.simple import estimate_tokens` | 1.0MB | ~21.63% | English text, minimum memory |
+| **Basic** | `from skimtoken.basic import estimate_tokens` | 0.9MB | ~27.05% | General use |
+| **Multilingual** | `from skimtoken.multilingual import estimate_tokens` | 0.9MB | ~15.93% | Non-English, mixed languages |
+| **Multilingual Simple** | `from skimtoken.multilingual_simple import estimate_tokens` | 0.9MB | ~15.11% | Fast multilingual estimation |
 
 ```python
 # Example: Choose method based on your needs
@@ -259,7 +235,7 @@ A: Beta = breaking changes possible.
 We are actively working to improve skimtoken's accuracy and performance:
 
 1. **Better estimation algorithms**: Moving beyond simple character multiplication to more sophisticated statistical models
-2. **Performance optimization**: Fixing the 60x slowdown in multilingual method
+2. **Performance optimization**: Further improving execution speed
 3. **Improved language support**: Better handling of non-English languages
 4. **Higher accuracy**: Targeting <10% error rate while maintaining low memory footprint
 

{skimtoken-0.2.1 → skimtoken-0.2.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "skimtoken"
-version = "0.2.1"
+version = "0.2.2"
 description = "Fast token count estimation library"
 readme = "README.md"
 requires-python = ">=3.9"

{skimtoken-0.2.1 → skimtoken-0.2.2}/scripts/update_rust_params.py

@@ -4,10 +4,11 @@ Script to load parameters from TOML files and update the Rust parameter files.
 Handles all parameter types: simple, basic, multilingual, and multilingual_simple.
 """
 
-import toml
 from pathlib import Path
 from typing import Any, Callable, TypedDict
 
+import toml
+
 
 class ParamConfig(TypedDict):
     name: str
@@ -23,6 +24,29 @@ def load_params_from_toml(toml_path: str) -> dict[str, Any]:
     return data
 
 
+def format_f32(value: float) -> str:
+    """Format a float value for f32 with underscores for readability."""
+    str_val = f"{value:.7g}"  # Use 7 significant digits for f32 precision
+
+    if "e" in str_val.lower():
+        return str_val
+
+    if "." in str_val:
+        integer_part, decimal_part = str_val.split(".")
+
+        if len(decimal_part) > 3:
+            formatted_decimal = ""
+            for i, digit in enumerate(decimal_part):
+                if i > 0 and i % 3 == 0:
+                    formatted_decimal += "_"
+                formatted_decimal += digit
+            return f"{integer_part}.{formatted_decimal}"
+        else:
+            return str_val
+    else:
+        return str_val
+
+
 def generate_simple_params_rust(params_data: dict[str, Any]) -> str:
     """Generate Rust code for simple parameters."""
     rust_code: list[str] = []
@@ -30,7 +54,7 @@ def generate_simple_params_rust(params_data: dict[str, Any]) -> str:
     rust_code.append("impl Default for SimpleParameters {")
     rust_code.append("    fn default() -> Self {")
     rust_code.append("        Self {")
-    rust_code.append(f"            coefficient: {params_data['coefficient']},")
+    rust_code.append(f"            coefficient: {format_f32(params_data['coefficient'])},")
     rust_code.append("        }")
     rust_code.append("    }")
     rust_code.append("}")
@@ -45,11 +69,13 @@ def generate_basic_params_rust(params_data: dict[str, Any]) -> str:
     rust_code.append("impl Default for BasicParameters {")
     rust_code.append("    fn default() -> Self {")
     rust_code.append("        Self {")
-    rust_code.append(f"            char_coef: {params_data['char_coef']},")
-    rust_code.append(f"            word_coef: {params_data['word_coef']},")
-    rust_code.append(f"            avg_word_length_coef: {params_data['avg_word_length_coef']},")
-    rust_code.append(f"            space_coef: {params_data['space_coef']},")
-    rust_code.append(f"            intercept: {params_data['intercept']},")
+    rust_code.append(f"            char_coef: {format_f32(params_data['char_coef'])},")
+    rust_code.append(f"            word_coef: {format_f32(params_data['word_coef'])},")
+    rust_code.append(
+        f"            avg_word_length_coef: {format_f32(params_data['avg_word_length_coef'])},"
+    )
+    rust_code.append(f"            space_coef: {format_f32(params_data['space_coef'])},")
+    rust_code.append(f"            intercept: {format_f32(params_data['intercept'])},")
     rust_code.append("        }")
     rust_code.append("    }")
     rust_code.append("}")
@@ -66,11 +92,13 @@ def generate_multilingual_params_rust(params_data: dict[str, Any]) -> str:
     rust_code.append("impl Default for MultilingualParameters {")
     rust_code.append("    fn default() -> Self {")
     rust_code.append("        Self {")
-    rust_code.append(f"            char_coef: {default_params['char_coef']},")
-    rust_code.append(f"            word_coef: {default_params['word_coef']},")
-    rust_code.append(f"            avg_word_length_coef: {default_params['avg_word_length_coef']},")
-    rust_code.append(f"            space_coef: {default_params['space_coef']},")
-    rust_code.append(f"            intercept: {default_params['intercept']},")
+    rust_code.append(f"            char_coef: {format_f32(default_params['char_coef'])},")
+    rust_code.append(f"            word_coef: {format_f32(default_params['word_coef'])},")
+    rust_code.append(
+        f"            avg_word_length_coef: {format_f32(default_params['avg_word_length_coef'])},"
+    )
+    rust_code.append(f"            space_coef: {format_f32(default_params['space_coef'])},")
+    rust_code.append(f"            intercept: {format_f32(default_params['intercept'])},")
     rust_code.append("        }")
     rust_code.append("    }")
     rust_code.append("}")
@@ -89,13 +117,13 @@ def generate_multilingual_params_rust(params_data: dict[str, Any]) -> str:
         rust_code.append("        language_params.insert(")
         rust_code.append(f'            "{lang_key}".to_string(),')
         rust_code.append("            MultilingualParameters {")
-        rust_code.append(f"                char_coef: {lang_params['char_coef']},")
-        rust_code.append(f"                word_coef: {lang_params['word_coef']},")
+        rust_code.append(f"                char_coef: {format_f32(lang_params['char_coef'])},")
+        rust_code.append(f"                word_coef: {format_f32(lang_params['word_coef'])},")
         rust_code.append(
-            f"                avg_word_length_coef: {lang_params['avg_word_length_coef']},"
+            f"                avg_word_length_coef: {format_f32(lang_params['avg_word_length_coef'])},"
         )
-        rust_code.append(f"                space_coef: {lang_params['space_coef']},")
-        rust_code.append(f"                intercept: {lang_params['intercept']},")
+        rust_code.append(f"                space_coef: {format_f32(lang_params['space_coef'])},")
+        rust_code.append(f"                intercept: {format_f32(lang_params['intercept'])},")
         rust_code.append("            },")
         rust_code.append("        );")
         rust_code.append("")
@@ -119,7 +147,7 @@ def generate_multilingual_simple_params_rust(params_data: dict[str, Any]) -> str
     rust_code.append("impl Default for MultilingualSimpleParameters {")
     rust_code.append("    fn default() -> Self {")
     rust_code.append("        Self {")
-    rust_code.append(f"            coefficient: {default_params['coefficient']},")
+    rust_code.append(f"            coefficient: {format_f32(default_params['coefficient'])},")
     rust_code.append("        }")
     rust_code.append("    }")
     rust_code.append("}")
@@ -137,7 +165,7 @@ def generate_multilingual_simple_params_rust(params_data: dict[str, Any]) -> str
         rust_code.append("        language_params.insert(")
         rust_code.append(f'            "{lang_key}".to_string(),')
         rust_code.append("            MultilingualSimpleParameters {")
-        rust_code.append(f"                coefficient: {lang_params['coefficient']},")
+        rust_code.append(f"                coefficient: {format_f32(lang_params['coefficient'])},")
         rust_code.append("            },")
         rust_code.append("        );")
         rust_code.append("")

{skimtoken-0.2.1 → skimtoken-0.2.2}/src/lib.rs

@@ -133,7 +133,7 @@ fn _skimtoken_core(m: &Bound<'_, PyModule>) -> PyResult<()> {
 
     #[pyfn(m)]
     #[pyo3(name = "count_basic")]
-    fn count_basic_py(text: &Bound<'_, PyAny>) -> PyResult<(usize, usize, f64, usize)> {
+    fn count_basic_py(text: &Bound<'_, PyAny>) -> PyResult<(usize, usize, f32, usize)> {
         let text_str = if let Ok(s) = text.extract::<String>() {
             s
         } else {
@@ -153,7 +153,7 @@ fn _skimtoken_core(m: &Bound<'_, PyModule>) -> PyResult<()> {
     #[pyo3(name = "count_multilingual")]
     fn count_multilingual_py(
         text: &Bound<'_, PyAny>,
-    ) -> PyResult<(usize, usize, f64, usize, String)> {
+    ) -> PyResult<(usize, usize, f32, usize, String)> {
         let text_str = if let Ok(s) = text.extract::<String>() {
             s
         } else {

{skimtoken-0.2.1 → skimtoken-0.2.2}/src/methods/method_basic.rs

@@ -5,27 +5,27 @@ use serde::{Deserialize, Serialize};
 pub struct BasicFeatures {
     pub char_count: usize,
     pub word_count: usize,
-    pub avg_word_length: f64,
+    pub avg_word_length: f32,
     pub space_count: usize,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct BasicParameters {
-    pub char_coef: f64,
-    pub word_coef: f64,
-    pub avg_word_length_coef: f64,
-    pub space_coef: f64,
-    pub intercept: f64,
+    pub char_coef: f32,
+    pub word_coef: f32,
+    pub avg_word_length_coef: f32,
+    pub space_coef: f32,
+    pub intercept: f32,
 }
 
 impl Default for BasicParameters {
     fn default() -> Self {
         Self {
-            char_coef: 0.3217745347518016,
-            word_coef: 0.07022881669049061,
-            avg_word_length_coef: 0.5090982427870748,
-            space_coef: -0.15831091236345404,
-            intercept: 1.591021053665763,
+            char_coef: 0.321_774_5,
+            word_coef: 0.070_228_82,
+            avg_word_length_coef: 0.509_098_2,
+            space_coef: -0.158_310_9,
+            intercept: 1.591_021,
         }
     }
 }
@@ -60,7 +60,7 @@ impl EstimationMethod for BasicMethod {
 
         let avg_word_length = if word_count > 0 {
             let total_word_chars: usize = words.iter().map(|w| w.chars().count()).sum();
-            total_word_chars as f64 / word_count as f64
+            total_word_chars as f32 / word_count as f32
         } else {
             0.0
         };
@@ -75,10 +75,10 @@ impl EstimationMethod for BasicMethod {
 
     fn estimate(&self, text: &str) -> usize {
         let features = self.count(text);
-        let estimate = self.parameters.char_coef * features.char_count as f64
-            + self.parameters.word_coef * features.word_count as f64
+        let estimate = self.parameters.char_coef * features.char_count as f32
+            + self.parameters.word_coef * features.word_count as f32
             + self.parameters.avg_word_length_coef * features.avg_word_length
-            + self.parameters.space_coef * features.space_count as f64
+            + self.parameters.space_coef * features.space_count as f32
             + self.parameters.intercept;
 
         estimate.round().max(0.0) as usize