xy-scale 1.1.2 → 1.1.4

package/README.md

@@ -28,7 +28,6 @@ This function prepares a dataset for supervised learning by parsing, scaling, an
 - `yCallbackFunc` (Function): Custom function to parse Y for each object. Return null or undefined to exclude it from training.
 - `xCallbackFunc` (Function): Custom function to parse X for each object. Return null or undefined to exclude it from training.
 - `forceScaling` (String, optional): Forces a specific scaling approach for each feature.
-- `timeSteps` (Number, optional): Transforms a one-dimensional array into an array of overlapping sequences (timesteps), each of a specified length. Default is 0 returning original output.
 
 #### Features:
 - **Y and X Parsing**: Custom parsing for Y and X based on user-defined functions.
@@ -57,7 +56,6 @@ Designed for production environments, this function parses and scales feature da
 - `weights` (Object, optional): Feature weights for scaling.
 - `xCallbackFunc` (Function): Custom function to parse X for each object. Return null or undefined to exclude it from production.
 - `forceScaling` (String, optional): Forces a specific scaling approach for each feature.
-- `timeSteps` (Number, optional): Transforms a one-dimensional array into an array of overlapping sequences (timesteps), each of a specified length. Default is 0 returning original output.
 
 #### Returns:
 - `X`: Scaled feature array for production data.
@@ -86,12 +84,28 @@ The `descaleArrayObj` function reverses the scaling applied to a dataset, recons
 - `scaled`: scaled array returned from `parseTrainingXY` (`trainX`, `trainY`, `testX` or `testY`) or `parseProductionX` (`X`).
 - `config`: config object returned from `parseTrainingXY` (`configX`, `configY`) or `parseProductionX` (`configX`).
 - `keyNames`: keyNames array returned from `parseTrainingXY` (`keyNamesX`, `keyNamesY`) or `parseProductionX` (`keyNamesX`).
-- `timeSteps`: input numeric timeSteps param submited to `parseTrainingXY` or `parseProductionX`. Defaults to 0. 
 
 #### Returns:
 
 The output of this function in an array of objects similar or equal to the input array of objects submited to `parseTrainingXY` or `parseProductionX`.
 
+### 4. `arrayToTimesteps`
+
+The `arrayToTimesteps` function transforms a flat array into an array of overlapping sequences of a specified length. This is a common preprocessing step for time-series or sequential data models.
+
+#### Parameters:
+- `arr` (Array): The input array to be converted into timesteps.
+- `timeSteps` (Number): The number of elements in each timestep sequence.
+  - Must be greater than `0`.
+  - If `timeSteps === 0`, the function returns the original array unchanged.
+
+#### Returns:
+An array of sub-arrays, where each sub-array contains `timeSteps` consecutive elements from the original array.
+
+#### Throws:
+- An error if `timeSteps < 0`, with the message: `"timeSteps must be greater than 0"`.
+
+
 ---
 
 ## Usage Examples
@@ -99,27 +113,32 @@ The output of this function in an array of objects similar or equal to the input
 1. **Parsing and Splitting a Training Dataset:**
 
 ```javascript
-    import { parseTrainingXY } from './scale.js';
-
-    const myArray = [
-        { open: 135.23, high: 137.45, low: 134.56, sma_200: 125.34, sma_100: 130.56 },
-        { open: 136.45, high: 138.67, low: 135.67, sma_200: 126.78, sma_100: 131.45 },
-        { open: 137.89, high: 139.34, low: 136.34, sma_200: 127.56, sma_100: 132.78 }
-    ];
+    import { parseTrainingXY, arrayToTimesteps } from 'xy-scale';
+    import { loadFile } from "./test/fs.js" 
+    import * as tf from '@tensorflow/tfjs-node';
 
+    // [{open, high, low, close}, {open, high, low, close}]
+    const myArray = await loadFile({fileName: '1d-spy.json', pathName: 'datasets'}) //file in /datasets/1d-spy.json
+    
+    //callback function used to prepare X before scaling
     const xCallbackFunc = ({ objRow, index }) => {
-        const curr = objRow[index];
-        const { open, high, low, sma_200, sma_100 } = curr;
+        const curr = objRow[index]
+        const prev = objRow[index - 1]
 
-        return {
-            open,
-            high,
-            low,
-            sma_200,
-            sma_100
-        };
-    };
+        //returning null or undefined will exclude current row X and Y from training
+        if(typeof prev === 'undefined') return null
+
+        const { open, high, low, close } = curr
 
+        return {
+            change: open - prev.close,
+            top: high - Math.max(open, close),
+            bottom: low - Math.min(open, close),
+            body: open-close,
+        }
+    }
+
+    //callback function used to prepare Y before scaling
     const yCallbackFunc = ({ objRow, index }) => {
         const curr = objRow[index];
         const next = objRow[index + 1];
@@ -128,15 +147,12 @@ The output of this function in an array of objects similar or equal to the input
         if (typeof next === 'undefined') return null;
 
         return {
-            label_1: next.open > curr.open,       // Label indicating if the next open price is higher than the current
-            label_2: next.high > curr.high,       // Label indicating if the next high price is higher than the current
-            label_3: next.low > curr.low,         // Label indicating if the next low price is higher than the current
-            label_4: next.sma_200 > curr.sma_200, // Label indicating if the next 200-day SMA is higher than the current
-            label_5: next.sma_100 > curr.sma_100  // Label indicating if the next 100-day SMA is higher than the current
+            label_1: next.open > curr.close,
+            label_2: next.close > curr.close,
         };
     };
-
-    const const {
+    
+    const {
         trainX,
         trainY,
         testX,
@@ -145,89 +161,32 @@ The output of this function in an array of objects similar or equal to the input
         keyNamesX,
     } = parseTrainingXY({
         arrObj: myArray,
-        trainingSplit: 0.75,
-        weights: { open: 1, high: 1, low: 1, sma_200: 1, sma_100: 1 },
+        trainingSplit: 0.90,
         yCallbackFunc,
         xCallbackFunc,
         forceScaling: 'normalization',
-        timeSteps: 0
     });
 
-    
-```
 
+    console.log('testX', testX.slice(-2))
 
-2. **Parsing a Production Dataset:**
 
-```javascript
-    import { parseProductionX } from './scale.js'
-    import {descaleArrayObj} from './descale.js'
+    const timeSteps = 10
+    const colsX = trainX[0].length
+    const colsY = trainY[0].length
+    const timeSteppedTrainX = arrayToTimesteps(trainX, timeSteps)
+    const trimedTrainY = trainY.slice(timeSteps-1)
 
-    const xCallbackFunc = ({ objRow, index }) => {
-        const curr = objRow[index];
-        const { open, high, low, sma_200, sma_100 } = curr;
 
-        return {
-            open,
-            high,
-            low,
-            sma_200,
-            sma_100
-        };
-    };
-
-    const myArray = [
-        { open: 135.23, high: 137.45, low: 134.56, sma_200: 125.34, sma_100: 130.56 },
-        { open: 136.45, high: 138.67, low: 135.67, sma_200: 126.78, sma_100: 131.45 },
-        { open: 137.89, high: 139.34, low: 136.34, sma_200: 127.56, sma_100: 132.78 }
-    ];
-
-    const {X, configX, keyNamesX} = parseProductionX({
-        arrObj: myArray,
-        weights: { open: 2, high: 1, low: 1, sma_200: 1, sma_100: 1 },
-        xCallbackFunc,
-        forceScaling: null,
-        timeSteps: 0
-    });
-
-    console.log('productionData.X', productionData.X)
-    console.log('descaled array be equal to myArray', descaleArrayObj({scaled: X, config: configX, keyNames: keyNamesX}))
-
-```
-
----
+    console.log([trainX.length, timeSteps, timeSteppedTrainX[0][0].length])
 
-### Upcoming Feature: Optional Precision Handling with Big.js and BigNumber.js
+    const inputX = tf.tensor3d(timeSteppedTrainX, [timeSteppedTrainX.length, timeSteps, colsX])
+    const targetY = tf.tensor2d(trimedTrainY,  [trimedTrainY.length, colsY])
 
-In the next release, we are introducing an optional **precision** feature to enhance decimal precision in financial and scientific datasets. This feature will allow users to integrate **Big.js** or **BigNumber.js** libraries seamlessly into their data processing workflow by adding a new `precision` property to the parameters of `parseTrainingXY` and `parseProductionX`.
 
-#### How Precision Handling Will Work
-
-With the new `precision` property, users can pass either Big.js or BigNumber.js as callback functions to handle high-precision decimal calculations. This makes the integration fully optional, allowing flexibility based on the precision requirements of the dataset. When `precision` is set, the toolkit will use the specified library for all numeric computations, ensuring high precision and minimizing rounding errors.
-
-1. **Future Example Usage:**
-
-```javascript
-    import Big from 'big.js';
-    import BigNumber from 'bignumber.js';
-    import { parseTrainingXY, parseProductionX } from './scale.js';
-
-    const myArray = [
-        { open: 135.23, high: 137.45, low: 134.56, sma_200: 125.34, sma_100: 130.56 },
-        { open: 136.45, high: 138.67, low: 135.67, sma_200: 126.78, sma_100: 131.45 },
-        { open: 137.89, high: 139.34, low: 136.34, sma_200: 127.56, sma_100: 132.78 }
-    ];
-
-    const trainingData = parseTrainingXY({
-        arrObj: myArray,
-        trainingSplit: 0.75,
-        weights: { open: 1, high: 1, low: 1, sma_200: 1, sma_100: 1 },
-        yCallbackFunc,
-        xCallbackFunc,
-        precision: Big, // Big or BigNumber callbacks for high-precision calculations
-        forceScaling: 'normalization',
-        timeSteps: 0
-    });
+    console.log('inputX', inputX)
+    console.log('inputX', targetY)
+    
 ```
 
 ---

package/dist/xy-scale.min.js

@@ -1 +1 @@
-var XY_Scale;(()=>{"use strict";var e={d:(t,n)=>{for(var r in n)e.o(n,r)&&!e.o(t,r)&&Object.defineProperty(t,r,{enumerable:!0,get:n[r]})},o:(e,t)=>Object.prototype.hasOwnProperty.call(e,t),r:e=>{"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})}},t={};e.r(t),e.d(t,{arrayToTimesteps:()=>s,descaleArrayObj:()=>a,parseProductionX:()=>o,parseTrainingXY:()=>r,timeStepsToArray:()=>i});const n=({arrObj:e,weights:t={},forceScaling:n=null})=>{if(null!==n&&"normalization"!==n&&"standardization"!==n)throw Error('forceScaling should be null, "normalization" or "standardization"');const r=e.length;if(0===r)return{scaledOutput:[],scaledConfig:{},keyNames:[]};const o=Object.keys(e[0]),a=o.map((e=>t.hasOwnProperty(e)?Math.max(t[e],1):1)),s=a.reduce(((e,t)=>e+t),0),i=new Array(s);let l=0;for(let e=0;e<o.length;e++)for(let t=0;t<a[e];t++)i[l++]=o[e];const c={},f={},u={},d={},g={},p={},h={},m={};for(const t of o){const n=e[0][t];c[t]=typeof n,"string"===c[t]&&(h[t]={}),f[t]=1/0,u[t]=-1/0,d[t]=0,g[t]=0,m[t]=0}for(const t of e)for(const e of o){let n=t[e];if("string"===c[e]){const r=h[e];r.hasOwnProperty(n)||(r[n]=Object.keys(r).length),n=r[n],t[e]=n}f[e]=Math.min(f[e],n),u[e]=Math.max(u[e],n),m[e]++;const r=n-d[e];d[e]+=r/m[e],g[e]+=r*(n-d[e])}const y={};for(const e of o)y[e]=m[e]>1?Math.sqrt(g[e]/(m[e]-1)):0,"normalization"===n||"standardization"===n?p[e]=n:0===f[e]&&1===u[e]?p[e]="none":p[e]=y[e]<1?"normalization":"standardization";const b=new Array(r);for(let t=0;t<r;t++){const n=e[t],r=new Array(s);let i=0;for(let e=0;e<o.length;e++){const t=o[e],s=n[t],l=f[t],c=u[t],g=d[t],h=y[t];let m;m="none"===p[t]?s:"normalization"===p[t]?c!==l?(s-l)/(c-l):0:0!==h?(s-g)/h:0;const b=a[e];for(let e=0;e<b;e++)r[i++]=m}b[t]=r}return{scaledOutput:b,scaledConfig:{min:f,max:u,std:y,mean:d,approach:p,inputTypes:c,uniqueStringIndexes:h},scaledKeyNames:i}},r=({arrObj:e,trainingSplit:t=.8,weights:r={},yCallbackFunc:o,xCallbackFunc:a,forceScaling:s,timeSteps:i=0})=>{const l=[],c=[];for(let t=0;t<e.length;t++){const n=a({objRow:e,index:t}),r=o({objRow:e,index:t});null!=n&&null!=r&&(l.push(n),c.push(r))}const{scaledOutput:f,scaledConfig:u,scaledKeyNames:d}=n({arrObj:l,weights:r,forceScaling:s}),{scaledOutput:g,scaledConfig:p,scaledKeyNames:h}=n({arrObj:c,weights:r,forceScaling:s}),m=Math.floor(f.length*t);return{trainX:f.slice(0,m),trainY:g.slice(0,m),testX:f.slice(m),testY:g.slice(m),configX:u,keyNamesX:d,configY:p,keyNamesY:h}},o=({arrObj:e,weights:t={},xCallbackFunc:r,forceScaling:o,timeSteps:a=0})=>{const s=[];for(let t=0;t<e.length;t++){const n=r({objRow:e,index:t});n&&s.push(n)}const{scaledOutput:i,scaledConfig:l,scaledKeyNames:c}=n({arrObj:s,weights:t,forceScaling:o});return{X:i,configX:l,keyNamesX:c}},a=({scaled:e,config:t,keyNames:n})=>{const{min:r,max:o,std:a,mean:s,approach:i,inputTypes:l,uniqueStringIndexes:c}=t;return e.map((e=>{const t={};let f=0;for(const u of Object.keys(r)){const d=i[u],g=r[u],p=o[u],h=s[u],m=a[u],y=n.filter((e=>e===u)).length;let b=0;for(let t=0;t<y;t++)b+=e[f++];const O=b/y;let S;if("normalization"===d?S=O*(p-g)+g:"standardization"===d&&(S=O*m+h),"string"===l[u]){const e=Object.keys(c[u]).find((e=>c[u][e]===S));S=void 0!==e?e:S}t[u]=S}return t}))},s=(e,t)=>{if(0===t)return e;if(t<0)throw new Error("timeSteps must be greater than 0");const n=[];for(let r=0;r<=e.length-t;r++)n.push(e.slice(r,r+t));return n},i=(e,t)=>{if(0===t)return e;if(t<0)throw new Error("timeSteps must be greater than 0");const n=[];n.push(...e[0]);for(let r=1;r<e.length;r++)n.push(e[r][t-1]);return n};XY_Scale=t})();
+var XY_Scale;(()=>{"use strict";var e={d:(t,n)=>{for(var o in n)e.o(n,o)&&!e.o(t,o)&&Object.defineProperty(t,o,{enumerable:!0,get:n[o]})},o:(e,t)=>Object.prototype.hasOwnProperty.call(e,t),r:e=>{"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})}},t={};e.r(t),e.d(t,{arrayToTimesteps:()=>s,descaleArrayObj:()=>a,parseProductionX:()=>r,parseTrainingXY:()=>o});const n=({arrObj:e,weights:t={},forceScaling:n=null})=>{if(null!==n&&"normalization"!==n&&"standardization"!==n)throw Error('forceScaling should be null, "normalization" or "standardization"');const o=e.length;if(0===o)return{scaledOutput:[],scaledConfig:{},keyNames:[]};const r=Object.keys(e[0]),a=r.map((e=>t.hasOwnProperty(e)?Math.max(t[e],1):1)),s=a.reduce(((e,t)=>e+t),0),i=new Array(s);let l=0;for(let e=0;e<r.length;e++)for(let t=0;t<a[e];t++)i[l++]=r[e];const c={},f={},d={},u={},g={},p={},m={},h={};for(const t of r){const n=e[0][t];c[t]=typeof n,"string"===c[t]&&(m[t]={}),f[t]=1/0,d[t]=-1/0,u[t]=0,g[t]=0,h[t]=0}for(const t of e)for(const e of r){let n=t[e];if("string"===c[e]){const o=m[e];o.hasOwnProperty(n)||(o[n]=Object.keys(o).length),n=o[n],t[e]=n}f[e]=Math.min(f[e],n),d[e]=Math.max(d[e],n),h[e]++;const o=n-u[e];u[e]+=o/h[e],g[e]+=o*(n-u[e])}const y={};for(const e of r)y[e]=h[e]>1?Math.sqrt(g[e]/(h[e]-1)):0,"normalization"===n||"standardization"===n?p[e]=n:0===f[e]&&1===d[e]?p[e]="none":p[e]=y[e]<1?"normalization":"standardization";const b=new Array(o);for(let t=0;t<o;t++){const n=e[t],o=new Array(s);let i=0;for(let e=0;e<r.length;e++){const t=r[e],s=n[t],l=f[t],c=d[t],g=u[t],m=y[t];let h;h="none"===p[t]?s:"normalization"===p[t]?c!==l?(s-l)/(c-l):0:0!==m?(s-g)/m:0;const b=a[e];for(let e=0;e<b;e++)o[i++]=h}b[t]=o}return{scaledOutput:b,scaledConfig:{min:f,max:d,std:y,mean:u,approach:p,inputTypes:c,uniqueStringIndexes:m},scaledKeyNames:i}},o=({arrObj:e,trainingSplit:t=.8,weights:o={},yCallbackFunc:r,xCallbackFunc:a,forceScaling:s,timeSteps:i=0})=>{const l=[],c=[];for(let t=0;t<e.length;t++){const n=a({objRow:e,index:t}),o=r({objRow:e,index:t});null!=n&&null!=o&&(l.push(n),c.push(o))}const{scaledOutput:f,scaledConfig:d,scaledKeyNames:u}=n({arrObj:l,weights:o,forceScaling:s}),{scaledOutput:g,scaledConfig:p,scaledKeyNames:m}=n({arrObj:c,weights:o,forceScaling:s}),h=Math.floor(f.length*t);return{trainX:f.slice(0,h),trainY:g.slice(0,h),testX:f.slice(h),testY:g.slice(h),configX:d,keyNamesX:u,configY:p,keyNamesY:m}},r=({arrObj:e,weights:t={},xCallbackFunc:o,forceScaling:r,timeSteps:a=0})=>{const s=[];for(let t=0;t<e.length;t++){const n=o({objRow:e,index:t});n&&s.push(n)}const{scaledOutput:i,scaledConfig:l,scaledKeyNames:c}=n({arrObj:s,weights:t,forceScaling:r});return{X:i,configX:l,keyNamesX:c}},a=({scaled:e,config:t,keyNames:n})=>{const{min:o,max:r,std:a,mean:s,approach:i,inputTypes:l,uniqueStringIndexes:c}=t;return e.map((e=>{const t={};let f=0;for(const d of Object.keys(o)){const u=i[d],g=o[d],p=r[d],m=s[d],h=a[d],y=n.filter((e=>e===d)).length;let b=0;for(let t=0;t<y;t++)b+=e[f++];const O=b/y;let S;if("normalization"===u?S=O*(p-g)+g:"standardization"===u&&(S=O*h+m),"string"===l[d]){const e=Object.keys(c[d]).find((e=>c[d][e]===S));S=void 0!==e?e:S}t[d]=S}return t}))},s=(e,t)=>{if(0===t)return e;if(t<0)throw new Error("timeSteps must be greater than 0");const n=[];for(let o=0;o<=e.length-t;o++)n.push(e.slice(o,o+t));return n};XY_Scale=t})();

package/index.js

@@ -1,5 +1,5 @@
 import { parseTrainingXY, parseProductionX } from "./src/datasets.js"
 import { descaleArrayObj } from "./src/descale.js"
-import { timeStepsToArray, arrayToTimesteps } from "./src/timeSteps.js"
+import {arrayToTimesteps } from "./src/timeSteps.js"
 
-export { parseTrainingXY, parseProductionX, descaleArrayObj, timeStepsToArray, arrayToTimesteps }
+export { parseTrainingXY, parseProductionX, descaleArrayObj, arrayToTimesteps }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "xy-scale",
-  "version": "1.1.2",
+  "version": "1.1.4",
   "main": "./index.js",
   "type": "module",
   "scripts": {
@@ -11,6 +11,7 @@
   "license": "ISC",
   "description": "This repository contains a JavaScript module designed to facilitate the preprocessing of datasets for machine learning applications. The primary functionality of the module is to scale feature data using normalization or standardization methods, and to parse training and production datasets into appropriate formats for training models.",
   "devDependencies": {
+    "@tensorflow/tfjs-node": "^4.22.0",
     "webpack": "^5.96.1",
     "webpack-cli": "^5.1.4"
   },

package/src/timeSteps.js

@@ -9,22 +9,4 @@ export const arrayToTimesteps = (arr, timeSteps) => {
     }
     
     return timestepsArray;
-}
-
-
-export const timeStepsToArray = (arr, timeSteps) => {
-    if (timeSteps === 0) return arr;
-    if (timeSteps < 0) throw new Error("timeSteps must be greater than 0");
-
-    const result = [];
-
-    // Start by adding all elements of the first timestep
-    result.push(...arr[0]);
-
-    // For each subsequent timestep, add only the last element
-    for (let i = 1; i < arr.length; i++) {
-        result.push(arr[i][timeSteps - 1]);
-    }
-
-    return result;
-};
+}

package/test/test.js

@@ -1,35 +1,41 @@
-import { parseTrainingXY, parseProductionX } from "../src/datasets.js";
-import { descaleArrayObj } from "../src/descale.js";
-import { loadFile } from "./fs.js";
+import { parseTrainingXY } from "../src/datasets.js"
+import {arrayToTimesteps} from '../src/timeSteps.js'
+import { loadFile } from "./fs.js"
+import * as tf from '@tensorflow/tfjs-node'
 
 const test = async () => {
 
-    const timeSteps = 10
-    const myArray = (await loadFile({fileName: '1d-spy.json', pathName: 'datasets'}))
+    const myArray = (await loadFile({fileName: '1d-spy.json', pathName: 'datasets'})) //file in /datasets/1d-spy.json
     
+    //callback function used to prepare X before scaling
     const xCallbackFunc = ({ objRow, index }) => {
-        const curr = objRow[index];
-        const { open, high, low} = curr;
-    
+        const curr = objRow[index]
+        const prev = objRow[index - 1]
+
         //returning null or undefined will exclude current row X and Y from training
+        if(typeof prev === 'undefined') return null
+
+        const { open, high, low, close } = curr
+
         return {
-            open,
-            high,
-            low,
-        };
-    };
-    
+            change: open - prev.close,
+            top: high - Math.max(open, close),
+            bottom: low - Math.min(open, close),
+            body: open-close,
+        }
+    }
+
+    //callback function used to prepare Y before scaling
     const yCallbackFunc = ({ objRow, index }) => {
         const curr = objRow[index];
         const next = objRow[index + 1];
-    
-         //returning null or undefined will exclude current row X and Y from training
+
+        //returning null or undefined will exclude current row X and Y from training
         if (typeof next === 'undefined') return null;
-    
+
         return {
-            label_1: next.open > curr.open,       // Label indicating if the next open price is higher than the current
-            label_2: next.high > curr.high,       // Label indicating if the next high price is higher than the current
-            label_3: next.low > curr.low,         // Label indicating if the next low price is higher than the current
+            label_1: next.open > curr.close,
+            label_2: next.close > curr.close,
         };
     };
     
@@ -42,22 +48,31 @@ const test = async () => {
         keyNamesX,
     } = parseTrainingXY({
         arrObj: myArray,
-        trainingSplit: 0.75,
-        weights: { open: 1, high: 1, low: 1, sma_200: 1, sma_100: 1 },
+        trainingSplit: 0.90,
         yCallbackFunc,
         xCallbackFunc,
-        forceScaling: null,
-        timeSteps
+        forceScaling: 'normalization',
     });
 
-    //console.log(JSON.stringify(trainingData))
 
     console.log('testX', testX.slice(-2))
-    const descaled = descaleArrayObj({scaled: testX, config: configX, keyNames: keyNamesX, timeSteps})
 
-    //comparing descaled values with the last values of input myArray
-    console.log(descaled.slice(-2)) //last 2 values of descaled
-    console.log(myArray.slice(-3, -1).map(({ open, high, low }) => ({ open, high, low }))) //last 2 values of myArray -1 interval
+
+    const timeSteps = 10
+    const colsX = trainX[0].length
+    const colsY = trainY[0].length
+    const timeSteppedTrainX = arrayToTimesteps(trainX, timeSteps)
+    const trimedTrainY = trainY.slice(timeSteps-1)
+
+
+    console.log([trainX.length, timeSteps, timeSteppedTrainX[0][0].length])
+
+    const inputX = tf.tensor3d(timeSteppedTrainX, [timeSteppedTrainX.length, timeSteps, colsX])
+    const targetY = tf.tensor2d(trimedTrainY,  [trimedTrainY.length, colsY])
+
+
+    console.log('inputX', inputX)
+    console.log('inputX', targetY)
 }
 
 test()