Iris Dataset¶

In [1]:

# Uncomment the following line to install GPU version with Cupy backend
#!pip install 'sorix @ git+https://github.com/Mitchell-Mirano/sorix.git@main'

# Uncomment the following line to install GPU version with Cupy backend #!pip install 'sorix @ git+https://github.com/Mitchell-Mirano/sorix.git@main'

In [2]:

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from typing import Union

import sorix
from sorix.model_selection import train_test_split
from sorix.preprocessing import StandardScaler,MinMaxScaler,RobustScaler
from sorix.nn import Module
from sorix import tensor,Tensor
from sorix.nn import CrossEntropyLoss
from sorix.metrics import accuracy_score
from sorix.optim import RMSprop
from sorix.nn import Linear,ReLU
from sorix.metrics import confusion_matrix, classification_report
import seaborn as sns

import numpy as np import pandas as pd import matplotlib.pyplot as plt from typing import Union import sorix from sorix.model_selection import train_test_split from sorix.preprocessing import StandardScaler,MinMaxScaler,RobustScaler from sorix.nn import Module from sorix import tensor,Tensor from sorix.nn import CrossEntropyLoss from sorix.metrics import accuracy_score from sorix.optim import RMSprop from sorix.nn import Linear,ReLU from sorix.metrics import confusion_matrix, classification_report import seaborn as sns

In [3]:

device = 'cuda' if sorix.cuda.is_available() else "cpu"
device

device = 'cuda' if sorix.cuda.is_available() else "cpu" device

✅ GPU basic operation passed
✅ GPU available: NVIDIA GeForce RTX 4070 Laptop GPU
CUDA runtime version: 13000
CuPy version: 13.6.0

Out[3]:

'cuda'

In [4]:

df  = pd.read_csv("../data/Iris.csv")
df.head()

df = pd.read_csv("../data/Iris.csv") df.head()

Out[4]:

	Id	SepalLengthCm	SepalWidthCm	PetalLengthCm	PetalWidthCm	Species
0	1	5.1	3.5	1.4	0.2	Iris-setosa
1	2	4.9	3.0	1.4	0.2	Iris-setosa
2	3	4.7	3.2	1.3	0.2	Iris-setosa
3	4	4.6	3.1	1.5	0.2	Iris-setosa
4	5	5.0	3.6	1.4	0.2	Iris-setosa

In [5]:

labels = df["Species"].unique()

labels2id = {label: i for i, label in enumerate(labels)}
id2labels = {i: label for i, label in enumerate(labels)}

labels = df["Species"].unique() labels2id = {label: i for i, label in enumerate(labels)} id2labels = {i: label for i, label in enumerate(labels)}

In [6]:

df['labels'] = df['Species'].map(labels2id)
df.head()

df['labels'] = df['Species'].map(labels2id) df.head()

Out[6]:

	Id	SepalLengthCm	SepalWidthCm	PetalLengthCm	PetalWidthCm	Species	labels
0	1	5.1	3.5	1.4	0.2	Iris-setosa	0
1	2	4.9	3.0	1.4	0.2	Iris-setosa	0
2	3	4.7	3.2	1.3	0.2	Iris-setosa	0
3	4	4.6	3.1	1.5	0.2	Iris-setosa	0
4	5	5.0	3.6	1.4	0.2	Iris-setosa	0

In [7]:

df_train, df_test = train_test_split(df, test_size=0.2, random_state=42)

df_train, df_test = train_test_split(df, test_size=0.2, random_state=42)

In [8]:

features = ["SepalLengthCm", "SepalWidthCm", "PetalLengthCm", "PetalWidthCm"]
labels = ["labels"]

X_train = df_train[features]
y_train = df_train[labels]

X_test = df_test[features]
y_test = df_test[labels]

features = ["SepalLengthCm", "SepalWidthCm", "PetalLengthCm", "PetalWidthCm"] labels = ["labels"] X_train = df_train[features] y_train = df_train[labels] X_test = df_test[features] y_test = df_test[labels]

In [9]:

plt.scatter(X_train['SepalLengthCm'], X_train['SepalWidthCm'], c=y_train['labels'])

plt.scatter(X_train['SepalLengthCm'], X_train['SepalWidthCm'], c=y_train['labels'])

Out[9]:

<matplotlib.collections.PathCollection at 0x7f56b2b8c2f0>

In [10]:

scaler = StandardScaler()

scaler.fit(X_train)
X_train_scaled = scaler.transform(X_train)
X_test_scaled = scaler.transform(X_test)

scaler = StandardScaler() scaler.fit(X_train) X_train_scaled = scaler.transform(X_train) X_test_scaled = scaler.transform(X_test)

In [11]:

X_train_scaled.shape, y_train.shape, X_test_scaled.shape, y_test.shape

X_train_scaled.shape, y_train.shape, X_test_scaled.shape, y_test.shape

Out[11]:

((120, 4), (120, 1), (30, 4), (30, 1))

In [12]:

X_train_tensor = tensor(X_train_scaled).to(device)
Y_train_tensor = tensor(y_train).to(device)

X_test_tensor = tensor(X_test_scaled).to(device)
Y_test_tensor = tensor(y_test).to(device)

print(f"X_train shape: {X_train_tensor.shape}, device: {X_train_tensor.device}")
print(f"Y_train shape: {Y_train_tensor.shape}, device: {Y_train_tensor.device}")
print(f"X_test shape: {X_test_tensor.shape}, device: {X_test_tensor.device}")
print(f"Y_test shape: {Y_test_tensor.shape}, device: {Y_test_tensor.device}")

X_train_tensor = tensor(X_train_scaled).to(device) Y_train_tensor = tensor(y_train).to(device) X_test_tensor = tensor(X_test_scaled).to(device) Y_test_tensor = tensor(y_test).to(device) print(f"X_train shape: {X_train_tensor.shape}, device: {X_train_tensor.device}") print(f"Y_train shape: {Y_train_tensor.shape}, device: {Y_train_tensor.device}") print(f"X_test shape: {X_test_tensor.shape}, device: {X_test_tensor.device}") print(f"Y_test shape: {Y_test_tensor.shape}, device: {Y_test_tensor.device}")

X_train shape: sorix.Size([120, 4]), device: cuda:0
Y_train shape: sorix.Size([120, 1]), device: cuda:0
X_test shape: sorix.Size([30, 4]), device: cuda:0
Y_test shape: sorix.Size([30, 1]), device: cuda:0

In [13]:

class Net(Module):

    def __init__(self):
        super().__init__()
        self.fc1 = Linear(4, 8)
        self.fc2 = Linear(8, 4)
        self.fc3 = Linear(4, 3)
        self.relu = ReLU()

    def forward(self, x: tensor) -> Tensor:
        x = self.fc1(x)
        x = self.relu(x)
        x = self.fc2(x)
        x = self.relu(x)
        x = self.fc3(x)
        return x
    

net = Net().to(device)
criterion = CrossEntropyLoss()
optimizer = RMSprop(net.parameters(), lr=1e-2)

class Net(Module): def __init__(self): super().__init__() self.fc1 = Linear(4, 8) self.fc2 = Linear(8, 4) self.fc3 = Linear(4, 3) self.relu = ReLU() def forward(self, x: tensor) -> Tensor: x = self.fc1(x) x = self.relu(x) x = self.fc2(x) x = self.relu(x) x = self.fc3(x) return x net = Net().to(device) criterion = CrossEntropyLoss() optimizer = RMSprop(net.parameters(), lr=1e-2)

In [14]:

logits = net(X_train_tensor)
logits

logits = net(X_train_tensor) logits

Out[14]:

tensor([[ 3.18380770e-02,  5.91685631e-02, -7.86396504e-03],
        [ 3.13370197e+00,  5.74086691e+00, -7.45613334e-01],
        [-6.27524252e-02, -1.36923843e+00,  1.58636662e+00],
        [ 3.68812375e-03, -3.08446122e-01,  4.18230142e-01],
        [-2.65776979e-02, -2.79022078e-01,  2.54321886e-01],
        [ 2.34201072e+00,  4.24755254e+00, -5.85250038e-01],
        [ 8.82216808e-03,  2.42461865e-03,  1.40565934e-02],
        [ 8.13916013e-02, -1.05761437e+00,  1.66882102e+00],
        [-2.64715199e-02, -1.49514641e-01,  9.24898444e-02],
        [-5.54242181e-04, -1.86377704e-02, -4.84749159e-02],
        [ 3.60032830e-02, -8.28358602e-01,  1.21232245e+00],
        [ 2.06400820e+00,  3.68548434e+00, -5.50437986e-01],
        [ 2.88001819e+00,  5.23662302e+00, -7.55856685e-01],
        [ 2.21337046e+00,  3.97225913e+00, -5.78986114e-01],
        [ 3.18288505e+00,  5.80945838e+00, -6.86235603e-01],
        [ 1.45123381e-02, -5.24926676e-01,  7.39887747e-01],
        [ 2.46062169e-02, -1.00287913e+00,  1.40291056e+00],
        [-5.21423812e-02, -1.27879173e-01, -4.72920072e-02],
        [-5.65749397e-03, -3.95742734e-02,  3.27227844e-02],
        [ 2.88399195e-02, -9.18760118e-01,  1.30674497e+00],
        [ 2.19221991e+00,  3.93667524e+00, -5.17301622e-01],
        [ 8.97614835e-03, -5.72666380e-01,  7.83763085e-01],
        [ 2.29760228e+00,  4.15590958e+00, -5.26909164e-01],
        [ 2.39446310e-02, -8.61873039e-01,  1.21521530e+00],
        [-3.89442505e-02, -1.43634853e+00,  1.75593019e+00],
        [ 6.55337450e-02, -1.01456623e+00,  1.55805238e+00],
        [-5.04178076e-02, -6.63503494e-01,  7.00384973e-01],
        [ 4.32010985e-02, -1.16846247e+00,  1.68417142e+00],
        [ 1.77281707e+00,  3.15398165e+00, -4.43245852e-01],
        [ 2.02470363e+00,  3.62813623e+00, -4.86150842e-01],
        [ 3.13682903e+00,  5.67645254e+00, -7.30882150e-01],
        [ 4.04362152e+00,  7.43207193e+00, -7.80373135e-01],
        [-5.05755007e-02, -4.29726041e-01,  3.92407766e-01],
        [ 2.54596961e+00,  4.60337048e+00, -5.74363675e-01],
        [ 2.30134087e+00,  4.11463922e+00, -5.71982983e-01],
        [ 2.10389747e-02, -6.10826781e-01,  8.75143434e-01],
        [-9.43018398e-04, -4.67122438e-01,  6.11088082e-01],
        [ 2.78381905e+00,  5.05903683e+00, -6.53340586e-01],
        [ 2.97422700e+00,  5.40743424e+00, -6.47728008e-01],
        [ 3.95251834e+00,  7.24417637e+00, -8.03992697e-01],
        [-8.61208329e-03, -5.44252653e-01,  6.86330607e-01],
        [-1.79378373e-03, -4.57687607e-01,  5.95774957e-01],
        [-3.87303344e-02, -5.21625127e-01,  5.53716051e-01],
        [ 3.30825152e+00,  6.05562187e+00, -7.02709854e-01],
        [ 3.14721013e+00,  5.74610699e+00, -7.31635734e-01],
        [-9.77670537e-02, -2.30847597e-01, -8.92878186e-02],
        [-3.39432586e-02, -3.70475437e-01,  3.71288841e-01],
        [ 6.65223663e-03, -7.32668999e-01,  9.86244692e-01],
        [-4.89626813e-02, -3.82594034e-01,  3.35932843e-01],
        [ 6.60773547e-02, -1.46109646e+00,  2.14713684e+00],
        [-1.15041971e-02, -8.57852344e-02,  7.35272471e-02],
        [-5.57455970e-02, -1.26605881e+00,  1.47460515e+00],
        [-2.70143290e-02, -2.49851595e-01,  2.36320584e-01],
        [ 2.80564147e+00,  5.09444191e+00, -6.36834297e-01],
        [-9.94160128e-02, -1.16481896e+00,  1.19232601e+00],
        [ 8.81469045e-02,  1.41017470e-01, -8.47095075e-02],
        [ 2.64496766e+00,  4.79361724e+00, -6.51898193e-01],
        [ 2.70862079e+00,  4.91054523e+00, -6.09610678e-01],
        [ 2.91926941e+00,  5.33392086e+00, -6.28574545e-01],
        [-5.69323871e-02, -1.83996163e-01, -4.85782950e-02],
        [ 1.91095232e-02, -5.71609068e-01,  8.16979266e-01],
        [ 2.44987675e+00,  4.43411215e+00, -5.48691117e-01],
        [ 1.87454068e+00,  3.34866676e+00, -5.28320474e-01],
        [ 1.95499250e+00,  3.52732934e+00, -4.56857253e-01],
        [-1.39832846e-02, -3.81138219e-02, -1.24192564e-02],
        [ 2.60212352e+00,  4.71550175e+00, -6.14193012e-01],
        [-1.74164042e-03, -3.97214924e-03, -1.60025383e-03],
        [ 8.84677919e-03, -1.56237158e+00,  2.08487385e+00],
        [ 2.25639293e+00,  4.04456673e+00, -5.41551958e-01],
        [-2.13781899e-02, -1.49540828e-01,  1.23650843e-01],
        [ 1.83769871e-02, -4.92952115e-01,  7.11036506e-01],
        [ 4.01361073e+00,  7.36482486e+00, -8.13147441e-01],
        [ 5.43918374e-03, -5.39444581e-01,  7.27994249e-01],
        [-8.61208329e-03, -5.44252653e-01,  6.86330607e-01],
        [ 1.84389463e-01,  2.88705400e-01, -1.94539596e-01],
        [-2.61323801e-02, -1.82796475e-01,  1.51148944e-01],
        [ 1.42086436e-02, -1.02002889e+00,  1.38995515e+00],
        [ 1.62906834e-02, -5.67155660e-05,  1.75329364e-02],
        [ 3.15816858e+00,  5.78813885e+00, -6.55640220e-01],
        [-7.37799912e-02, -2.10869806e-01, -6.48543265e-02],
        [ 1.97121320e-02, -6.98977705e-01,  9.86538615e-01],
        [ 2.45276126e+00,  4.42872696e+00, -5.85644883e-01],
        [ 2.34624314e+00,  4.22006766e+00, -5.94801937e-01],
        [-4.58528401e-02, -1.23924859e-01, -4.07969126e-02],
        [-1.55761148e-02, -6.71523850e-01,  8.29920866e-01],
        [ 2.21337046e+00,  3.97225913e+00, -5.78986114e-01],
        [ 1.37889201e-02, -8.47158929e-01,  1.16118199e+00],
        [ 2.59161266e+00,  4.69414655e+00, -5.87087276e-01],
        [ 2.24495156e+00,  4.06386263e+00, -5.36941918e-01],
        [-5.86844805e-03, -5.57119991e-01,  7.12622161e-01],
        [-5.34681391e-02, -1.62438706e-01, -4.63365464e-02],
        [ 5.75974294e-02, -1.15981793e+00,  1.72196818e+00],
        [-7.66682471e-03, -3.74621636e-01,  4.66478519e-01],
        [-1.93847917e-02, -1.35596971e-01,  1.12121085e-01],
        [-1.66016791e-02, -1.43525842e+00,  1.83079905e+00],
        [ 1.07662092e-02, -2.58287947e-03,  1.17626312e-02],
        [ 1.71668511e+00,  3.02509832e+00, -4.54053266e-01],
        [ 2.12903590e+00,  3.77705174e+00, -5.44621435e-01],
        [-1.16346873e-02, -9.66011259e-02,  8.10755029e-02],
        [-4.86084197e-02, -9.53046868e-01,  1.08734018e+00],
        [ 2.04438529e+00,  3.65162275e+00, -4.99071343e-01],
        [ 2.66797028e+00,  4.84329065e+00, -5.88822143e-01],
        [ 1.93382898e+00,  3.42756963e+00, -4.99609979e-01],
        [-3.85869265e-02, -2.69916254e-01,  2.23185687e-01],
        [ 1.90211079e-02, -1.12409321e+00,  1.54324434e+00],
        [ 2.50570150e+00,  4.51384544e+00, -6.00848088e-01],
        [ 4.05301890e-02, -9.79425855e-01,  1.42644957e+00],
        [ 1.15126805e-02, -1.21833120e+00,  1.64153374e+00],
        [ 2.28844858e+00,  4.10618547e+00, -5.43055515e-01],
        [-1.89513001e-02, -2.51221781e-01,  2.65658676e-01],
        [-4.04414381e-02, -4.00419444e-01,  3.88475909e-01],
        [ 2.39035689e-02, -5.41035522e-01,  7.93144484e-01],
        [-7.62890172e-02, -5.34633177e-01,  4.42555595e-01],
        [ 3.62928176e-02, -1.00766438e+00,  1.44911469e+00],
        [ 3.26473531e+00,  5.96281259e+00, -6.88166431e-01],
        [ 4.13973561e-02, -1.01511886e+00,  1.47635059e+00],
        [ 5.31529751e-03, -2.65784310e-01,  3.67682955e-01],
        [ 3.65343726e-02, -8.05372287e-01,  1.18390708e+00],
        [-8.60919474e-02, -2.00252294e-01, -8.98781550e-02],
        [-2.30541384e-02, -6.82638008e-02, -2.01015647e-02]], device='cuda:0', dtype=sorix.float64, requires_grad=True)

In [15]:

# Bucle de entrenamiento mejorado
for epoch in range(1000 + 1):
    logits = net(X_train_tensor)
    loss = criterion(logits, Y_train_tensor)

    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

    if epoch % 10 == 0:
        preds = sorix.argmax(logits, axis=1, keepdims=True)
        acc_train = (preds== Y_train_tensor).mean()
        with sorix.no_grad():
            logits = net(X_test_tensor)
            preds = sorix.argmax(logits, axis=1, keepdims=True)
            acc_test = (preds == Y_test_tensor).mean()

        # Usamos una f-string para formatear y alinear la salida
        print(f"[{device}] Epoch {epoch:5d} | Loss: {loss.item():.4f} | Acc Train: {acc_train.item()*100:.2f}% | Acc Test: {acc_test.item()*100:.2f}%")

        if acc_test.item() >= 0.96:  # Mejoramos el criterio de parada
            print(f"Entrenamiento completado en {epoch} epochs!")
            break

# Bucle de entrenamiento mejorado for epoch in range(1000 + 1): logits = net(X_train_tensor) loss = criterion(logits, Y_train_tensor) optimizer.zero_grad() loss.backward() optimizer.step() if epoch % 10 == 0: preds = sorix.argmax(logits, axis=1, keepdims=True) acc_train = (preds== Y_train_tensor).mean() with sorix.no_grad(): logits = net(X_test_tensor) preds = sorix.argmax(logits, axis=1, keepdims=True) acc_test = (preds == Y_test_tensor).mean() # Usamos una f-string para formatear y alinear la salida print(f"[{device}] Epoch {epoch:5d} | Loss: {loss.item():.4f} | Acc Train: {acc_train.item()*100:.2f}% | Acc Test: {acc_test.item()*100:.2f}%") if acc_test.item() >= 0.96: # Mejoramos el criterio de parada print(f"Entrenamiento completado en {epoch} epochs!") break

[cuda] Epoch     0 | Loss: 1.3640 | Acc Train: 34.17% | Acc Test: 63.33%
[cuda] Epoch    10 | Loss: 0.2004 | Acc Train: 93.33% | Acc Test: 96.67%
Entrenamiento completado en 10 epochs!

In [16]:

with sorix.no_grad():
    logits = net(X_test_tensor)
    preds = sorix.argmax(logits, axis=1, keepdims=True)

with sorix.no_grad(): logits = net(X_test_tensor) preds = sorix.argmax(logits, axis=1, keepdims=True)

In [17]:

sns.heatmap(confusion_matrix(Y_test_tensor, preds), annot=True, cmap="Blues")

sns.heatmap(confusion_matrix(Y_test_tensor, preds), annot=True, cmap="Blues")

Out[17]:

<Axes: >

In [18]:

print(classification_report(Y_test_tensor, preds))

print(classification_report(Y_test_tensor, preds))

            precision   recall f1-score  support
0                1.00     1.00     1.00        7
1                0.92     1.00     0.96       11
2                1.00     0.92     0.96       12

accuracy                           0.97       30
macro avg        0.97     0.97     0.97       30
weighted avg     0.97     0.97     0.97       30