Refactor training loop from script to class

README.md

@@ -1 +1,99 @@
-# evals
+# evals
+
+This project provides a Python implementation for training and evaluating a simple neural network on the MNIST dataset using PyTorch.
+
+## MNISTTrainer class
+
+The `MNISTTrainer` class is used to train and evaluate the model. It is defined in `src/main.py`.
+
+### Usage
+
+First, create an instance of the `MNISTTrainer` class:
+
+```python
+trainer = MNISTTrainer()
+```
+
+You can then train the model using the `train` method:
+
+```python
+trainer.train()
+```
+
+The trained model's parameters are automatically saved to a file named "mnist_model.pth".
+
+To load the model parameters from this file, use the following code:
+
+```python
+trainer.model.load_state_dict(torch.load("mnist_model.pth"))
+```
+
+To evaluate the model, you can use the `predict` method in `src/api.py`. This method takes an image file as input and returns the model's prediction.
+
+## Dependencies
+
+This project requires the following Python libraries:
+
+- PyTorch
+- torchvision
+- numpy
+- PIL
+- FastAPI
+
+Before installing these dependencies, ensure that `pip` is installed and working correctly. You can verify this by running the following command:
+
+```bash
+pip --version
+```
+
+If `pip` is not installed or not working correctly, you may need to troubleshoot your Python installation or install `pip` separately.
+
+Once `pip` is working correctly, you can install the `poetry` package manager using `pip`:
+
+```bash
+pip install poetry
+```
+
+After installing `poetry`, verify that it was installed correctly by checking its version:
+
+```bash
+poetry --version
+```
+
+If the command does not return a version number, `poetry` was not installed correctly.
+
+If you encounter issues with the `poetry` installation, you may need to troubleshoot your `poetry` installation. Here are some steps you can take:
+
+1. Check if the `poetry` executable is in your system's PATH. You can do this by running the following command:
+
+```bash
+which poetry
+```
+
+If the `poetry` executable is in your PATH, this command will print its location. If it's not, it won't print anything.
+
+2. If the `poetry` executable is not in your PATH, you need to add it. The process for this varies depending on your operating system and shell, but generally involves adding a line to a shell startup file like `~/.bashrc` or `~/.bash_profile` that exports the `poetry` executable's location to the PATH. You can find more detailed instructions in the `poetry` documentation or by searching online for "add to PATH".
+
+3. If you're still having trouble, you can seek help from the `poetry` community or check the `poetry` documentation for more troubleshooting tips. Here's the link to the `poetry` documentation: https://python-poetry.org/docs/
+
+Once `poetry` is installed, you can install the project dependencies using `poetry`:
+
+```bash
+poetry install
+```
+
+If you are unable to install `poetry`, you can install the project dependencies directly using `pip`:
+
+```bash
+pip install torch torchvision numpy pillow fastapi
+```
+
+## Running the project
+
+To run the project, first start the FastAPI server:
+
+```bash
+uvicorn src.api:app --reload
+```
+
+You can then send a POST request to the `/predict` endpoint with an image file to get the model's prediction.

src/api.py

@@ -1,19 +1,36 @@
+import os
+import subprocess
+import sys
+
+# Check if pip is installed
+try:
+    subprocess.run(["pip", "--version"], check=True)
+except subprocess.CalledProcessError:
+    print("Error: pip is not installed. Please install pip and try again.")
+    sys.exit(1)
+
+# Check if poetry is installed, if not, install it
+try:
+    subprocess.run(["poetry", "--version"], check=True)
+except subprocess.CalledProcessError:
+    # Download the get-poetry.py script
+    subprocess.run(["curl", "-sSL", "https://install.python-poetry.org", "-o", "get-poetry.py"], check=True)
+    # Execute the get-poetry.py script to install poetry
+    subprocess.run(["python", "get-poetry.py", "--yes"], check=True)
+
 from fastapi import FastAPI, UploadFile, File
 from PIL import Image
 import torch
 from torchvision import transforms
-from main import Net  # Importing Net class from main.py
+from main import MNISTTrainer  # Importing MNISTTrainer class from main.py
 
-# Load the model
-model = Net()
-model.load_state_dict(torch.load("mnist_model.pth"))
-model.eval()
+# Create an instance of MNISTTrainer and load the model
+trainer = MNISTTrainer()
+trainer.model.load_state_dict(torch.load("mnist_model.pth"))
+trainer.model.eval()
 
 # Transform used for preprocessing the image
-transform = transforms.Compose([
-    transforms.ToTensor(),
-    transforms.Normalize((0.5,), (0.5,))
-])
+transform = trainer.transform
 
 app = FastAPI()
 
@@ -23,6 +40,6 @@ async def predict(file: UploadFile = File(...)):
     image = transform(image)
     image = image.unsqueeze(0)  # Add batch dimension
     with torch.no_grad():
-        output = model(image)
+        output = trainer.model(image)
         _, predicted = torch.max(output.data, 1)
     return {"prediction": int(predicted[0])}

src/main.py

@@ -6,43 +6,49 @@
 from torch.utils.data import DataLoader
 import numpy as np
 
-# Step 1: Load MNIST Data and Preprocess
-transform = transforms.Compose([
-    transforms.ToTensor(),
-    transforms.Normalize((0.5,), (0.5,))
-])
+class MNISTTrainer:
+    def __init__(self, epochs=3):
+        self.transform = transforms.Compose([
+            transforms.ToTensor(),
+            transforms.Normalize((0.5,), (0.5,))
+        ])
+        self.trainset = datasets.MNIST('.', download=True, train=True, transform=self.transform)
+        self.trainloader = DataLoader(self.trainset, batch_size=64, shuffle=True)
+        self.testset = datasets.MNIST('.', download=True, train=False, transform=self.transform)
+        self.testloader = DataLoader(self.testset, batch_size=64, shuffle=True)
+        self.model = self._define_model()
+        self.optimizer = optim.SGD(self.model.parameters(), lr=0.01)
+        self.criterion = nn.NLLLoss()
+        self.epochs = epochs
 
-trainset = datasets.MNIST('.', download=True, train=True, transform=transform)
-trainloader = DataLoader(trainset, batch_size=64, shuffle=True)
+    def _define_model(self):
+        class Net(nn.Module):
+            def __init__(self):
+                super().__init__()
+                self.fc1 = nn.Linear(28 * 28, 128)
+                self.fc2 = nn.Linear(128, 64)
+                self.fc3 = nn.Linear(64, 10)
 
-# Step 2: Define the PyTorch Model
-class Net(nn.Module):
-    def __init__(self):
-        super().__init__()
-        self.fc1 = nn.Linear(28 * 28, 128)
-        self.fc2 = nn.Linear(128, 64)
-        self.fc3 = nn.Linear(64, 10)
-
-    def forward(self, x):
-        x = x.view(-1, 28 * 28)
-        x = nn.functional.relu(self.fc1(x))
-        x = nn.functional.relu(self.fc2(x))
-        x = self.fc3(x)
-        return nn.functional.log_softmax(x, dim=1)
+            def forward(self, x):
+                x = x.view(-1, 28 * 28)
+                x = nn.functional.relu(self.fc1(x))
+                x = nn.functional.relu(self.fc2(x))
+                x = self.fc3(x)
+                return nn.functional.log_softmax(x, dim=1)
+        return Net()
 
-# Step 3: Train the Model
-model = Net()
-optimizer = optim.SGD(model.parameters(), lr=0.01)
-criterion = nn.NLLLoss()
+    def train(self):
+        for epoch in range(self.epochs):
+            for images, labels in self.trainloader:
+                self.optimizer.zero_grad()
+                output = self.model(images)
+                loss = self.criterion(output, labels)
+                loss.backward()
+                self.optimizer.step()
 
-# Training loop
-epochs = 3
-for epoch in range(epochs):
-    for images, labels in trainloader:
-        optimizer.zero_grad()
-        output = model(images)
-        loss = criterion(output, labels)
-        loss.backward()
-        optimizer.step()
+        torch.save(self.model.state_dict(), "mnist_model.pth")
+
+trainer = MNISTTrainer()
+trainer.train()
 
 torch.save(model.state_dict(), "mnist_model.pth")