I'd like to save all the model parameters (optimizer, learning rate, batch size, etc.) and model architecture (number and types of layers) alongside of the model so that later go back analyze why some models works better.
Is there a simple way to store this metadata along with the weights?
From the docs:
from keras.models import load_model
model.save('my_model.h5') # creates a HDF5 file 'my_model.h5'
del model # deletes the existing model
# returns a compiled model
# identical to the previous one
model = load_model('my_model.h5')
This includes the optimizer (which should include the learning rate and the batch size). Besides that, you can use
Using configuration scripts which are under version control (I did so for my masters thesis, see my configuration scripts)
Storing the training script
If you want one file, just use a container file format like .tar.
Related
In DL4J, is there a way to access the columns of preprocessed data after at the scoring step? I have a case where I have a csv of data that includes fields that are not used by the neural net for prediction, but they are important to include in my output after the predictions are made.
Before I train my model, I have been using this step for preprocessing to remove the columns prior to training:
TransformProcess transformProcess =
new transformProcess.Builder(schema).removeColumns(columnsToOmit).build());
RecordReader transformProcessRecordReader =
new TransformProcessRecordReader(recordReader, transformProcess);
The problem I am running into is that after this transformation, I can of course train or make predictions, but I can no longer access those columns that are removed.
Is there a way to "ignore" the columns rather than removing them so that I can access them after the model makes a prediction?
In my debugger I can see some protected fields that show the data is still there but i'm really trying to avoid a custom implementation of iterator if there is a simpler way to do this.
I am quite new to nvidia-tlt. Currently, I have trained, pruned and retrained the model with the kitti dataset, also am able to do these steps on any datasets with the required kitti format. What I want to do is used a previously trained model on kitti and fine tune it to my own data. The config file have the options pretrained_model_path, resume_model_path and pruned_model_path, So there is no option for the fine-tune in config. If I try to use pretrained_model_path, it throws an exception for the shape.
Invalid argument: Incompatible shapes: [6,29484,3] vs. [6,29484,12]
That error is expected.
Technically the pretrained model that we download from ngc comes without final layer which represents the total number of classes and their respective bboxes.
Once you train that model with any dataset, then the trained model will be frozen with the top layer. Now, if you want to finetune the same model with different number of classes you will get error related to invalid shapes.
You need to train the model on the new dataset from the beginning.
If you want to finetune the model with different dataset but of the same classes then you can use the previously trained model.
I want to build a image classifier i gathered images from web and i resized them using PIL libray
now i want those images to be converted as input .what operations do i need to perform on these
images.I also did covert images in to numpy arrays and stored them in an list named features and what to do next
Well there are a number of decisions to make. One is to partition your images into a training set, a validation set and generally also a test set. I typically use 10% of the images as a validation set and 10% of the images as a test set. Next you need to decide how you want to provide your images to the network. My preference is to use the Keras ImageDataGenerator.flow from directory. This requires you to create 3 directories to store images. I put the test images in a directory called 'test', the validation images in a directory called 'valid' and the training images in a directory called 'train'. Now in each of these directories you need to create identically named class directories. For example if you are trying to classify images of dogs and cats. You would create a 'dogs' sub directory and 'cats' sub directory within the test, train and valid directories. Be sure to name them identically because the names of the sub directories determine the names of your classes. Now populate the class directories with your images. These can be images in standard formats like jpg. Now create 3 generators a train generator, a validation generator and a test generator as in
train_gen=ImageDataGenerator(preprocessing_function=pre_process).flow_from_directory('train', target_size=(height, width), batch_size=train_batch_size, seed=rand_seed, class_mode='categorical', color_mode='rgb')
do the same for the validation generator and the test generator. Documentation for the ImageDataGenerator and flow_from_directory is here.. Now you have your images stored and the data generators set up to provide data to your model in batches based on batch size. So now we can get to actually building a model. You can build your own model however there are excellent models for image processing available for you to use. This is called transfer learning. I like to use a model called MobileNet. I prefer this because it has a small number of trainable parameters (about 4 million) versus other models which have 10's of millions. Keras has this and many other image processing models . Documentation is here. Now you have to modify the final layer of the model to adapt it to your application. MobileNet was trained on the ImageNet data set that had 1000 classes. You need to remove this last layer and make it a dense layer having as many nodes as you have classes and use the softmax activation function. An example for the case of 2 classes is shown below.
mobile = tf.keras.applications.mobilenet.MobileNet( include_top=Top,
input_shape=(height,width,3),
pooling='avg', weights='imagenet',
alpha=1, depth_multiplier=1)
x=mobile.layers[-2].output
predictions=Dense (2, activation='softmax')(x)
model = Model(inputs=mobile.input, outputs=predictions)
for layer in model.layers:
layer.trainable=True
model.compile(Adam(lr=.001, loss='categorical_crossentropy', metrics=['accuracy'])
The last line of code compiles your model using the Adam optimizer with a learning rate of .001.Now we can finally get to training the model. I use the modelfit generator as shown below:
data = model.fit_generator(generator = train_gen,validation_data=val_gen, epochs=epochs, initial_epoch=start_epoch,
callbacks = callbacks, verbose=1)
Documentation for the above is here. The model will train on your training set and validate on the validation set. For each epoch (training cycle) you will get a print out of the training loss, training accuracy, validation loss and validation accuracy so you can monitor how your model is performing. The final step is to run your test set to see how well your model performs on data it was not trained on. Do do that use the code below:
resultspmodel.evaluate(test_gen, verbose=0)
print('Model accuracy on Test Set is {0:7.2f} %'.format(results[1]* 100)
That's about it but of course there are a lot of details to fill in. If you are new to Convolutional Neural Networks and machine learning I would recommend an excellent tutorial on YouTube at here. There are about 20 sequential tutorials in the play list. I used this tutorial as a beginner and found it excellent. It will cover all the topics you need to become skilled at using CNN classifiers. Good Luck!
How train_on_batch() is different from fit()? What are the cases when we should use train_on_batch()?
For this question, it's a simple answer from the primary author:
With fit_generator, you can use a generator for the validation data as
well. In general, I would recommend using fit_generator, but using
train_on_batch works fine too. These methods only exist for the sake of
convenience in different use cases, there is no "correct" method.
train_on_batch allows you to expressly update weights based on a collection of samples you provide, without regard to any fixed batch size. You would use this in cases when that is what you want: to train on an explicit collection of samples. You could use that approach to maintain your own iteration over multiple batches of a traditional training set but allowing fit or fit_generator to iterate batches for you is likely simpler.
One case when it might be nice to use train_on_batch is for updating a pre-trained model on a single new batch of samples. Suppose you've already trained and deployed a model, and sometime later you've received a new set of training samples previously never used. You could use train_on_batch to directly update the existing model only on those samples. Other methods can do this too, but it is rather explicit to use train_on_batch for this case.
Apart from special cases like this (either where you have some pedagogical reason to maintain your own cursor across different training batches, or else for some type of semi-online training update on a special batch), it is probably better to just always use fit (for data that fits in memory) or fit_generator (for streaming batches of data as a generator).
train_on_batch() gives you greater control of the state of the LSTM, for example, when using a stateful LSTM and controlling calls to model.reset_states() is needed. You may have multi-series data and need to reset the state after each series, which you can do with train_on_batch(), but if you used .fit() then the network would be trained on all the series of data without resetting the state. There's no right or wrong, it depends on what data you're using, and how you want the network to behave.
Train_on_batch will also see a performance increase over fit and fit generator if youre using large datasets and don't have easily serializable data (like high rank numpy arrays), to write to tfrecords.
In this case you can save the arrays as numpy files and load up smaller subsets of them (traina.npy, trainb.npy etc) in memory, when the whole set won't fit in memory. You can then use tf.data.Dataset.from_tensor_slices and then using train_on_batch with your subdataset, then loading up another dataset and calling train on batch again, etc, now you've trained on your entire set and can control exactly how much and what of your dataset trains your model. You can then define your own epochs, batch sizes, etc with simple loops and functions to grab from your dataset.
Indeed #nbro answer helps, just to add few more scenarios, lets say you are training some seq to seq model or a large network with one or more encoders. We can create custom training loops using train_on_batch and use a part of our data to validate on the encoder directly without using callbacks. Writing callbacks for a complex validation process could be difficult. There are several cases where we wish to train on batch.
Regards,
Karthick
From Keras - Model training APIs:
fit: Trains the model for a fixed number of epochs (iterations on a dataset).
train_on_batch: Runs a single gradient update on a single batch of data.
We can use it in GAN when we update the discriminator and generator using a batch of our training data set at a time. I saw Jason Brownlee used train_on_batch in on his tutorials (How to Develop a 1D Generative Adversarial Network From Scratch in Keras)
Tip for quick search: Type Control+F and type in the search box the term that you want to search (train_on_batch, for example).
I am trying to export my model to Google Cloud Storage. I used tf.contrib.learn to build my model and followed the iris classification example.
After my training and evaluation is done I would like to store the model on the cloud so I can make predictions, but I don't know how to export the model.
classifier = tf.contrib.learn.DNNClassifier(feature_columns=feature_columns,
hidden_units=[100],
n_classes=50,
model_dir="Model_Logs")
The best example of training on the cloud is probably census (canned estimator) or census (custom estimator). They use the same Estimator API, so that part should be familiar. In addition, they use the Estimator class to help perform the training automatically. The train_and_evaluate method is called on that class by learn_runner.run, which will export the model if properly configured, which basically boils down to setting the export_strategy and the model_dir
If you want to do things outside the Experiment and learn_runner frameworks, you can just call Estimator.export_savedmodel