Update sample function in browser and node env

Author: risenW

danfojs-browser/src/core/series.js

@@ -14,7 +14,7 @@
 */
 
 
-import { tensor, round } from "@tensorflow/tfjs";
+import * as tf from "@tensorflow/tfjs";
 import { variance, std } from 'mathjs';
 import { Utils } from "./utils";
 import { Str } from "./strings";
@@ -55,7 +55,7 @@ export class Series extends NDframe {
     * @returns {1D Tensor}
     */
   get tensor() {
-    return tensor(this.values).asType(this.dtypes[0]);
+    return tf.tensor(this.values).asType(this.dtypes[0]);
   }
 
 
@@ -95,30 +95,22 @@ export class Series extends NDframe {
   }
 
   /**
-    * Returns n number of random rows in a Series
-    * @param {rows} number of rows to return
-    * @returns {Series}
+   * Gets [num] number of random rows in a dataframe
+   * @param {num}  rows --> The number of rows to return
+   * @param {seed}  seed --> (Optional) An integer specifying the random seed that will be used to create the distribution.
+   * @returns {Promise} resolves to a Series object
     */
-  sample(num = 5) {
-    if (num > this.values.length || num < 1) {
-      let config = { columns: this.column_names };
-      return new Series(this.values, config);
-    } else {
-      let values = this.values;
-      let idx = this.index;
-      let new_values = [];
-      let new_idx = [];
-      let rand_nums = utils.__shuffle(num, idx);
-
-      rand_nums.forEach((i) => {
-        new_values.push(values[i]);
-        new_idx.push(idx[i]);
-      });
-      let config = { columns: this.column_names, index: new_idx };
-      let sf = new Series(new_values, config);
-      return sf;
-
+  async sample(num = 5, seed = 1) {
+    if (num > this.shape[0]) {
+      throw new Error("Sample size n cannot be bigger than size of dataset");
     }
+    if (num < -1 || num == 0) {
+      throw new Error("Sample size cannot be less than -1 or 0");
+    }
+    num = num === -1 ? this.shape[0] : num;
+    const shuffled_index = await tf.data.array(this.index).shuffle(num, seed).take(num).toArray();
+    const sf = this.iloc(shuffled_index);
+    return sf;
   }
 
   /**
@@ -250,7 +242,7 @@ export class Series extends NDframe {
   mean() {
     utils._throw_str_dtype_error(this, 'mean');
     let values = utils._remove_nans(this.values);
-    let mean = tensor(values).mean().arraySync();
+    let mean = tf.tensor(values).mean().arraySync();
     return mean;
   }
 
@@ -382,7 +374,7 @@ export class Series extends NDframe {
   round(dp) {
     if (utils.__is_undefined(dp)) {
       //use tensorflow round function to roound to the nearest whole number
-      let result = round(this.row_data_tensor).arraySync();
+      let result = tf.round(this.row_data_tensor).arraySync();
       return new Series(result, { columns: this.column_names, index: this.index });
 
     } else {

danfojs-browser/tests/core/series.js

@@ -1,5 +1,5 @@
 /* eslint-disable no-undef */
-const tf = require("@tensorflow/tfjs-core");
+const tf = require("@tensorflow/tfjs");
 
 describe("Series", function () {
   describe("tensor", function () {
@@ -78,20 +78,25 @@ describe("Series", function () {
   });
 
   describe("sample", function () {
-    it("Samples n number of random elements from a DataFrame", function () {
+    it("Samples n number of random elements from a DataFrame", async function () {
       let data = [ 1, 2, 3, 4, 5, 620, 30, 40, 39, 89, 78 ];
       let sf = new dfd.Series(data);
-      assert.deepEqual(sf.sample(7).values.length, 7);
+      assert.deepEqual((await sf.sample(7)).values.length, 7);
     });
-    it("Return all values if n of sample is greater than lenght of Dataframe", function () {
+    it("Return all values if n of sample -1", async function () {
       let data = [ 1, 2, 3, 4, 5, 620, 30, 40, 39, 89, 78 ];
       let sf = new dfd.Series(data);
-      assert.deepEqual(sf.sample(21).values.length, data.length);
+      assert.deepEqual((await sf.sample(-1)).values.length, data.length);
     });
-    it("Return all values if n of sample is less than 1", function () {
+    it("Throw error if n is greater than lenght of Series", async function () {
       let data = [ 1, 2, 3, 4, 5, 620, 30, 40, 39, 89, 78 ];
       let sf = new dfd.Series(data);
-      assert.deepEqual(sf.sample(-2).values.length, data.length);
+      try {
+        await sf.sample(100);
+      } catch (e) {
+        expect(e).to.be.instanceOf(Error);
+        expect(e.message).to.eql('Sample size n cannot be bigger than size of dataset');
+      }
     });
   });
 

danfojs-node/src/core/series.js

@@ -94,30 +94,22 @@ export class Series extends NDframe {
   }
 
   /**
-    * Returns n number of random rows in a Series
-    * @param {rows} number of rows to return
-    * @returns {Series}
+   * Gets [num] number of random rows in a dataframe
+   * @param {num}  rows --> The number of rows to return
+   * @param {seed}  seed --> (Optional) An integer specifying the random seed that will be used to create the distribution.
+   * @returns {Promise} resolves to a Series object
     */
-  sample(num = 5) {
-    if (num > this.values.length || num < 1) {
-      let config = { columns: this.column_names };
-      return new Series(this.values, config);
-    } else {
-      let values = this.values;
-      let idx = this.index;
-      let new_values = [];
-      let new_idx = [];
-      let rand_nums = utils.__shuffle(num, idx);
-
-      rand_nums.forEach((i) => {
-        new_values.push(values[i]);
-        new_idx.push(idx[i]);
-      });
-      let config = { columns: this.column_names, index: new_idx };
-      let sf = new Series(new_values, config);
-      return sf;
-
+  async sample(num = 5, seed = 1) {
+    if (num > this.shape[0]) {
+      throw new Error("Sample size n cannot be bigger than size of dataset");
     }
+    if (num < -1 || num == 0) {
+      throw new Error("Sample size cannot be less than -1 or 0");
+    }
+    num = num === -1 ? this.shape[0] : num;
+    const shuffled_index = await tf.data.array(this.index).shuffle(num, seed).take(num).toArray();
+    const sf = this.iloc(shuffled_index);
+    return sf;
   }
 
   /**

danfojs-node/tests/core/series.js

@@ -1,4 +1,4 @@
-import { assert } from "chai";
+import { assert, expect } from "chai";
 import { Series } from "../../src/core/series";
 import * as tf from '@tensorflow/tfjs-node';
 
@@ -79,20 +79,25 @@ describe("Series", function () {
   });
 
   describe("sample", function () {
-    it("Samples n number of random elements from a DataFrame", function () {
+    it("Samples n number of random elements from a DataFrame", async function () {
       let data = [ 1, 2, 3, 4, 5, 620, 30, 40, 39, 89, 78 ];
       let sf = new Series(data);
-      assert.deepEqual(sf.sample(7).values.length, 7);
+      assert.deepEqual((await sf.sample(7)).values.length, 7);
     });
-    it("Return all values if n of sample is greater than lenght of Dataframe", function () {
+    it("Return all values if n of sample -1", async function () {
       let data = [ 1, 2, 3, 4, 5, 620, 30, 40, 39, 89, 78 ];
       let sf = new Series(data);
-      assert.deepEqual(sf.sample(21).values.length, data.length);
+      assert.deepEqual((await sf.sample(-1)).values.length, data.length);
     });
-    it("Return all values if n of sample is less than 1", function () {
+    it("Throw error if n is greater than lenght of Series", async function () {
       let data = [ 1, 2, 3, 4, 5, 620, 30, 40, 39, 89, 78 ];
       let sf = new Series(data);
-      assert.deepEqual(sf.sample(-2).values.length, data.length);
+      try {
+        await sf.sample(100);
+      } catch (e) {
+        expect(e).to.be.instanceOf(Error);
+        expect(e.message).to.eql('Sample size n cannot be bigger than size of dataset');
+      }
     });
   });